2 * Copyright (c) 2002 Luigi Rizzo, Universita` di Pisa
4 * Redistribution and use in source and binary forms, with or without
5 * modification, are permitted provided that the following conditions
7 * 1. Redistributions of source code must retain the above copyright
8 * notice, this list of conditions and the following disclaimer.
9 * 2. Redistributions in binary form must reproduce the above copyright
10 * notice, this list of conditions and the following disclaimer in the
11 * documentation and/or other materials provided with the distribution.
13 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
14 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
15 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
16 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
17 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
18 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
19 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
20 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
21 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
22 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * Implement IP packet firewall (new version)
35 #if !defined(KLD_MODULE)
37 #include "opt_ip6fw.h"
41 #error IPFIREWALL requires INET.
44 #include "opt_inet6.h"
45 #include "opt_ipsec.h"
47 #include <sys/param.h>
48 #include <sys/systm.h>
49 #include <sys/condvar.h>
50 #include <sys/malloc.h>
52 #include <sys/kernel.h>
55 #include <sys/module.h>
57 #include <sys/rwlock.h>
58 #include <sys/socket.h>
59 #include <sys/socketvar.h>
60 #include <sys/sysctl.h>
61 #include <sys/syslog.h>
62 #include <sys/ucred.h>
64 #include <net/radix.h>
65 #include <net/route.h>
66 #include <netinet/in.h>
67 #include <netinet/in_systm.h>
68 #include <netinet/in_var.h>
69 #include <netinet/in_pcb.h>
70 #include <netinet/ip.h>
71 #include <netinet/ip_var.h>
72 #include <netinet/ip_icmp.h>
73 #include <netinet/ip_fw.h>
74 #include <netinet/ip_divert.h>
75 #include <netinet/ip_dummynet.h>
76 #include <netinet/tcp.h>
77 #include <netinet/tcp_timer.h>
78 #include <netinet/tcp_var.h>
79 #include <netinet/tcpip.h>
80 #include <netinet/udp.h>
81 #include <netinet/udp_var.h>
83 #include <netgraph/ng_ipfw.h>
85 #include <altq/if_altq.h>
88 #include <netinet6/ipsec.h>
91 #include <netinet/ip6.h>
92 #include <netinet/icmp6.h>
94 #include <netinet6/scope6_var.h>
97 #include <netinet/if_ether.h> /* XXX for ETHERTYPE_IP */
99 #include <machine/in_cksum.h> /* XXX for in_cksum */
102 * set_disable contains one bit per set value (0..31).
103 * If the bit is set, all rules with the corresponding set
104 * are disabled. Set RESVD_SET(31) is reserved for the default rule
105 * and rules that are not deleted by the flush command,
106 * and CANNOT be disabled.
107 * Rules in set RESVD_SET can only be deleted explicitly.
109 static u_int32_t set_disable;
111 static int fw_verbose;
112 static int verbose_limit;
114 static struct callout ipfw_timeout;
115 static uma_zone_t ipfw_dyn_rule_zone;
116 #define IPFW_DEFAULT_RULE 65535
119 * Data structure to cache our ucred related
120 * information. This structure only gets used if
121 * the user specified UID/GID based constraints in
125 gid_t fw_groups[NGROUPS];
131 #define IPFW_TABLES_MAX 128
133 struct ip_fw *rules; /* list of rules */
134 struct ip_fw *reap; /* list of rules to reap */
135 struct radix_node_head *tables[IPFW_TABLES_MAX];
138 #define IPFW_LOCK_INIT(_chain) \
139 rw_init(&(_chain)->rwmtx, "IPFW static rules")
140 #define IPFW_LOCK_DESTROY(_chain) rw_destroy(&(_chain)->rwmtx)
141 #define IPFW_WLOCK_ASSERT(_chain) do { \
142 rw_assert(&(_chain)->rwmtx, RA_WLOCKED); \
143 NET_ASSERT_GIANT(); \
146 #define IPFW_RLOCK(p) rw_rlock(&(p)->rwmtx)
147 #define IPFW_RUNLOCK(p) rw_runlock(&(p)->rwmtx)
148 #define IPFW_WLOCK(p) rw_wlock(&(p)->rwmtx)
149 #define IPFW_WUNLOCK(p) rw_wunlock(&(p)->rwmtx)
152 * list of rules for layer 3
154 static struct ip_fw_chain layer3_chain;
156 MALLOC_DEFINE(M_IPFW, "IpFw/IpAcct", "IpFw/IpAcct chain's");
157 MALLOC_DEFINE(M_IPFW_TBL, "ipfw_tbl", "IpFw tables");
160 struct radix_node rn[2];
161 struct sockaddr_in addr, mask;
165 static int fw_debug = 1;
166 static int autoinc_step = 100; /* bounded to 1..1000 in add_rule() */
169 SYSCTL_NODE(_net_inet_ip, OID_AUTO, fw, CTLFLAG_RW, 0, "Firewall");
170 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, enable,
171 CTLFLAG_RW | CTLFLAG_SECURE3,
172 &fw_enable, 0, "Enable ipfw");
173 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, autoinc_step, CTLFLAG_RW,
174 &autoinc_step, 0, "Rule number autincrement step");
175 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, one_pass,
176 CTLFLAG_RW | CTLFLAG_SECURE3,
178 "Only do a single pass through ipfw when using dummynet(4)");
179 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, debug, CTLFLAG_RW,
180 &fw_debug, 0, "Enable printing of debug ip_fw statements");
181 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose,
182 CTLFLAG_RW | CTLFLAG_SECURE3,
183 &fw_verbose, 0, "Log matches to ipfw rules");
184 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose_limit, CTLFLAG_RW,
185 &verbose_limit, 0, "Set upper limit of matches of ipfw rules logged");
188 * Description of dynamic rules.
190 * Dynamic rules are stored in lists accessed through a hash table
191 * (ipfw_dyn_v) whose size is curr_dyn_buckets. This value can
192 * be modified through the sysctl variable dyn_buckets which is
193 * updated when the table becomes empty.
195 * XXX currently there is only one list, ipfw_dyn.
197 * When a packet is received, its address fields are first masked
198 * with the mask defined for the rule, then hashed, then matched
199 * against the entries in the corresponding list.
200 * Dynamic rules can be used for different purposes:
202 * + enforcing limits on the number of sessions;
203 * + in-kernel NAT (not implemented yet)
205 * The lifetime of dynamic rules is regulated by dyn_*_lifetime,
206 * measured in seconds and depending on the flags.
208 * The total number of dynamic rules is stored in dyn_count.
209 * The max number of dynamic rules is dyn_max. When we reach
210 * the maximum number of rules we do not create anymore. This is
211 * done to avoid consuming too much memory, but also too much
212 * time when searching on each packet (ideally, we should try instead
213 * to put a limit on the length of the list on each bucket...).
215 * Each dynamic rule holds a pointer to the parent ipfw rule so
216 * we know what action to perform. Dynamic rules are removed when
217 * the parent rule is deleted. XXX we should make them survive.
219 * There are some limitations with dynamic rules -- we do not
220 * obey the 'randomized match', and we do not do multiple
221 * passes through the firewall. XXX check the latter!!!
223 static ipfw_dyn_rule **ipfw_dyn_v = NULL;
224 static u_int32_t dyn_buckets = 256; /* must be power of 2 */
225 static u_int32_t curr_dyn_buckets = 256; /* must be power of 2 */
227 static struct mtx ipfw_dyn_mtx; /* mutex guarding dynamic rules */
228 #define IPFW_DYN_LOCK_INIT() \
229 mtx_init(&ipfw_dyn_mtx, "IPFW dynamic rules", NULL, MTX_DEF)
230 #define IPFW_DYN_LOCK_DESTROY() mtx_destroy(&ipfw_dyn_mtx)
231 #define IPFW_DYN_LOCK() mtx_lock(&ipfw_dyn_mtx)
232 #define IPFW_DYN_UNLOCK() mtx_unlock(&ipfw_dyn_mtx)
233 #define IPFW_DYN_LOCK_ASSERT() mtx_assert(&ipfw_dyn_mtx, MA_OWNED)
236 * Timeouts for various events in handing dynamic rules.
238 static u_int32_t dyn_ack_lifetime = 300;
239 static u_int32_t dyn_syn_lifetime = 20;
240 static u_int32_t dyn_fin_lifetime = 1;
241 static u_int32_t dyn_rst_lifetime = 1;
242 static u_int32_t dyn_udp_lifetime = 10;
243 static u_int32_t dyn_short_lifetime = 5;
246 * Keepalives are sent if dyn_keepalive is set. They are sent every
247 * dyn_keepalive_period seconds, in the last dyn_keepalive_interval
248 * seconds of lifetime of a rule.
249 * dyn_rst_lifetime and dyn_fin_lifetime should be strictly lower
250 * than dyn_keepalive_period.
253 static u_int32_t dyn_keepalive_interval = 20;
254 static u_int32_t dyn_keepalive_period = 5;
255 static u_int32_t dyn_keepalive = 1; /* do send keepalives */
257 static u_int32_t static_count; /* # of static rules */
258 static u_int32_t static_len; /* size in bytes of static rules */
259 static u_int32_t dyn_count; /* # of dynamic rules */
260 static u_int32_t dyn_max = 4096; /* max # of dynamic rules */
262 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_buckets, CTLFLAG_RW,
263 &dyn_buckets, 0, "Number of dyn. buckets");
264 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, curr_dyn_buckets, CTLFLAG_RD,
265 &curr_dyn_buckets, 0, "Current Number of dyn. buckets");
266 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_count, CTLFLAG_RD,
267 &dyn_count, 0, "Number of dyn. rules");
268 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_max, CTLFLAG_RW,
269 &dyn_max, 0, "Max number of dyn. rules");
270 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, static_count, CTLFLAG_RD,
271 &static_count, 0, "Number of static rules");
272 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_ack_lifetime, CTLFLAG_RW,
273 &dyn_ack_lifetime, 0, "Lifetime of dyn. rules for acks");
274 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_syn_lifetime, CTLFLAG_RW,
275 &dyn_syn_lifetime, 0, "Lifetime of dyn. rules for syn");
276 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_fin_lifetime, CTLFLAG_RW,
277 &dyn_fin_lifetime, 0, "Lifetime of dyn. rules for fin");
278 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_rst_lifetime, CTLFLAG_RW,
279 &dyn_rst_lifetime, 0, "Lifetime of dyn. rules for rst");
280 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_udp_lifetime, CTLFLAG_RW,
281 &dyn_udp_lifetime, 0, "Lifetime of dyn. rules for UDP");
282 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_short_lifetime, CTLFLAG_RW,
283 &dyn_short_lifetime, 0, "Lifetime of dyn. rules for other situations");
284 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_keepalive, CTLFLAG_RW,
285 &dyn_keepalive, 0, "Enable keepalives for dyn. rules");
289 * IPv6 specific variables
291 SYSCTL_DECL(_net_inet6_ip6);
293 static struct sysctl_ctx_list ip6_fw_sysctl_ctx;
294 static struct sysctl_oid *ip6_fw_sysctl_tree;
296 #endif /* SYSCTL_NODE */
298 static int fw_deny_unknown_exthdrs = 1;
302 * L3HDR maps an ipv4 pointer into a layer3 header pointer of type T
303 * Other macros just cast void * into the appropriate type
305 #define L3HDR(T, ip) ((T *)((u_int32_t *)(ip) + (ip)->ip_hl))
306 #define TCP(p) ((struct tcphdr *)(p))
307 #define UDP(p) ((struct udphdr *)(p))
308 #define ICMP(p) ((struct icmphdr *)(p))
309 #define ICMP6(p) ((struct icmp6_hdr *)(p))
312 icmptype_match(struct icmphdr *icmp, ipfw_insn_u32 *cmd)
314 int type = icmp->icmp_type;
316 return (type <= ICMP_MAXTYPE && (cmd->d[0] & (1<<type)) );
319 #define TT ( (1 << ICMP_ECHO) | (1 << ICMP_ROUTERSOLICIT) | \
320 (1 << ICMP_TSTAMP) | (1 << ICMP_IREQ) | (1 << ICMP_MASKREQ) )
323 is_icmp_query(struct icmphdr *icmp)
325 int type = icmp->icmp_type;
327 return (type <= ICMP_MAXTYPE && (TT & (1<<type)) );
332 * The following checks use two arrays of 8 or 16 bits to store the
333 * bits that we want set or clear, respectively. They are in the
334 * low and high half of cmd->arg1 or cmd->d[0].
336 * We scan options and store the bits we find set. We succeed if
338 * (want_set & ~bits) == 0 && (want_clear & ~bits) == want_clear
340 * The code is sometimes optimized not to store additional variables.
344 flags_match(ipfw_insn *cmd, u_int8_t bits)
349 if ( ((cmd->arg1 & 0xff) & bits) != 0)
350 return 0; /* some bits we want set were clear */
351 want_clear = (cmd->arg1 >> 8) & 0xff;
352 if ( (want_clear & bits) != want_clear)
353 return 0; /* some bits we want clear were set */
358 ipopts_match(struct ip *ip, ipfw_insn *cmd)
360 int optlen, bits = 0;
361 u_char *cp = (u_char *)(ip + 1);
362 int x = (ip->ip_hl << 2) - sizeof (struct ip);
364 for (; x > 0; x -= optlen, cp += optlen) {
365 int opt = cp[IPOPT_OPTVAL];
367 if (opt == IPOPT_EOL)
369 if (opt == IPOPT_NOP)
372 optlen = cp[IPOPT_OLEN];
373 if (optlen <= 0 || optlen > x)
374 return 0; /* invalid or truncated */
382 bits |= IP_FW_IPOPT_LSRR;
386 bits |= IP_FW_IPOPT_SSRR;
390 bits |= IP_FW_IPOPT_RR;
394 bits |= IP_FW_IPOPT_TS;
398 return (flags_match(cmd, bits));
402 tcpopts_match(struct tcphdr *tcp, ipfw_insn *cmd)
404 int optlen, bits = 0;
405 u_char *cp = (u_char *)(tcp + 1);
406 int x = (tcp->th_off << 2) - sizeof(struct tcphdr);
408 for (; x > 0; x -= optlen, cp += optlen) {
410 if (opt == TCPOPT_EOL)
412 if (opt == TCPOPT_NOP)
426 bits |= IP_FW_TCPOPT_MSS;
430 bits |= IP_FW_TCPOPT_WINDOW;
433 case TCPOPT_SACK_PERMITTED:
435 bits |= IP_FW_TCPOPT_SACK;
438 case TCPOPT_TIMESTAMP:
439 bits |= IP_FW_TCPOPT_TS;
444 return (flags_match(cmd, bits));
448 iface_match(struct ifnet *ifp, ipfw_insn_if *cmd)
450 if (ifp == NULL) /* no iface with this packet, match fails */
452 /* Check by name or by IP address */
453 if (cmd->name[0] != '\0') { /* match by name */
456 if (fnmatch(cmd->name, ifp->if_xname, 0) == 0)
459 if (strncmp(ifp->if_xname, cmd->name, IFNAMSIZ) == 0)
466 TAILQ_FOREACH(ia, &ifp->if_addrhead, ifa_link) {
467 if (ia->ifa_addr == NULL)
469 if (ia->ifa_addr->sa_family != AF_INET)
471 if (cmd->p.ip.s_addr == ((struct sockaddr_in *)
472 (ia->ifa_addr))->sin_addr.s_addr)
473 return(1); /* match */
476 return(0); /* no match, fail ... */
480 * The verify_path function checks if a route to the src exists and
481 * if it is reachable via ifp (when provided).
483 * The 'verrevpath' option checks that the interface that an IP packet
484 * arrives on is the same interface that traffic destined for the
485 * packet's source address would be routed out of. The 'versrcreach'
486 * option just checks that the source address is reachable via any route
487 * (except default) in the routing table. These two are a measure to block
488 * forged packets. This is also commonly known as "anti-spoofing" or Unicast
489 * Reverse Path Forwarding (Unicast RFP) in Cisco-ese. The name of the knobs
490 * is purposely reminiscent of the Cisco IOS command,
492 * ip verify unicast reverse-path
493 * ip verify unicast source reachable-via any
495 * which implements the same functionality. But note that syntax is
496 * misleading. The check may be performed on all IP packets whether unicast,
497 * multicast, or broadcast.
500 verify_path(struct in_addr src, struct ifnet *ifp)
503 struct sockaddr_in *dst;
505 bzero(&ro, sizeof(ro));
507 dst = (struct sockaddr_in *)&(ro.ro_dst);
508 dst->sin_family = AF_INET;
509 dst->sin_len = sizeof(*dst);
511 rtalloc_ign(&ro, RTF_CLONING);
513 if (ro.ro_rt == NULL)
517 * If ifp is provided, check for equality with rtentry.
518 * We should use rt->rt_ifa->ifa_ifp, instead of rt->rt_ifp,
519 * in order to pass packets injected back by if_simloop():
520 * if useloopback == 1 routing entry (via lo0) for our own address
521 * may exist, so we need to handle routing assymetry.
523 if (ifp != NULL && ro.ro_rt->rt_ifa->ifa_ifp != ifp) {
528 /* if no ifp provided, check if rtentry is not default route */
530 satosin(rt_key(ro.ro_rt))->sin_addr.s_addr == INADDR_ANY) {
535 /* or if this is a blackhole/reject route */
536 if (ifp == NULL && ro.ro_rt->rt_flags & (RTF_REJECT|RTF_BLACKHOLE)) {
541 /* found valid route */
548 * ipv6 specific rules here...
551 icmp6type_match (int type, ipfw_insn_u32 *cmd)
553 return (type <= ICMP6_MAXTYPE && (cmd->d[type/32] & (1<<(type%32)) ) );
557 flow6id_match( int curr_flow, ipfw_insn_u32 *cmd )
560 for (i=0; i <= cmd->o.arg1; ++i )
561 if (curr_flow == cmd->d[i] )
566 /* support for IP6_*_ME opcodes */
568 search_ip6_addr_net (struct in6_addr * ip6_addr)
572 struct in6_ifaddr *fdm;
573 struct in6_addr copia;
575 TAILQ_FOREACH(mdc, &ifnet, if_link)
576 for (mdc2 = mdc->if_addrlist.tqh_first; mdc2;
577 mdc2 = mdc2->ifa_list.tqe_next) {
580 if (mdc2->ifa_addr->sa_family == AF_INET6) {
581 fdm = (struct in6_ifaddr *)mdc2;
582 copia = fdm->ia_addr.sin6_addr;
583 /* need for leaving scope_id in the sock_addr */
584 in6_clearscope(&copia);
585 if (IN6_ARE_ADDR_EQUAL(ip6_addr, &copia))
593 verify_path6(struct in6_addr *src, struct ifnet *ifp)
596 struct sockaddr_in6 *dst;
598 bzero(&ro, sizeof(ro));
600 dst = (struct sockaddr_in6 * )&(ro.ro_dst);
601 dst->sin6_family = AF_INET6;
602 dst->sin6_len = sizeof(*dst);
603 dst->sin6_addr = *src;
604 rtalloc_ign((struct route *)&ro, RTF_CLONING);
606 if (ro.ro_rt == NULL)
610 * if ifp is provided, check for equality with rtentry
611 * We should use rt->rt_ifa->ifa_ifp, instead of rt->rt_ifp,
612 * to support the case of sending packets to an address of our own.
613 * (where the former interface is the first argument of if_simloop()
614 * (=ifp), the latter is lo0)
616 if (ifp != NULL && ro.ro_rt->rt_ifa->ifa_ifp != ifp) {
621 /* if no ifp provided, check if rtentry is not default route */
623 IN6_IS_ADDR_UNSPECIFIED(&satosin6(rt_key(ro.ro_rt))->sin6_addr)) {
628 /* or if this is a blackhole/reject route */
629 if (ifp == NULL && ro.ro_rt->rt_flags & (RTF_REJECT|RTF_BLACKHOLE)) {
634 /* found valid route */
640 hash_packet6(struct ipfw_flow_id *id)
643 i = (id->dst_ip6.__u6_addr.__u6_addr32[0]) ^
644 (id->dst_ip6.__u6_addr.__u6_addr32[1]) ^
645 (id->dst_ip6.__u6_addr.__u6_addr32[2]) ^
646 (id->dst_ip6.__u6_addr.__u6_addr32[3]) ^
647 (id->dst_port) ^ (id->src_port) ^ (id->flow_id6);
652 is_icmp6_query(int icmp6_type)
654 if ((icmp6_type <= ICMP6_MAXTYPE) &&
655 (icmp6_type == ICMP6_ECHO_REQUEST ||
656 icmp6_type == ICMP6_MEMBERSHIP_QUERY ||
657 icmp6_type == ICMP6_WRUREQUEST ||
658 icmp6_type == ICMP6_FQDN_QUERY ||
659 icmp6_type == ICMP6_NI_QUERY))
666 send_reject6(struct ip_fw_args *args, int code, u_short offset, u_int hlen)
668 if (code == ICMP6_UNREACH_RST && offset == 0 &&
669 args->f_id.proto == IPPROTO_TCP) {
679 if (args->m->m_len < (hlen+sizeof(struct tcphdr))) {
680 args->m = m_pullup(args->m, hlen+sizeof(struct tcphdr));
685 ip6 = mtod(args->m, struct ip6_hdr *);
686 tcp = (struct tcphdr *)(mtod(args->m, char *) + hlen);
688 if ((tcp->th_flags & TH_RST) != 0) {
695 ti.th.th_seq = ntohl(ti.th.th_seq);
696 ti.th.th_ack = ntohl(ti.th.th_ack);
697 ti.ip6.ip6_nxt = IPPROTO_TCP;
699 if (ti.th.th_flags & TH_ACK) {
705 if (((args->m)->m_flags & M_PKTHDR) != 0) {
706 ack += (args->m)->m_pkthdr.len - hlen
707 - (ti.th.th_off << 2);
708 } else if (ip6->ip6_plen) {
709 ack += ntohs(ip6->ip6_plen) + sizeof(*ip6)
710 - hlen - (ti.th.th_off << 2);
715 if (tcp->th_flags & TH_SYN)
718 flags = TH_RST|TH_ACK;
720 bcopy(&ti, ip6, sizeof(ti));
721 tcp_respond(NULL, ip6, (struct tcphdr *)(ip6 + 1),
722 args->m, ack, seq, flags);
724 } else if (code != ICMP6_UNREACH_RST) { /* Send an ICMPv6 unreach. */
725 icmp6_error(args->m, ICMP6_DST_UNREACH, code, 0);
735 static u_int64_t norule_counter; /* counter for ipfw_log(NULL...) */
737 #define SNPARGS(buf, len) buf + len, sizeof(buf) > len ? sizeof(buf) - len : 0
738 #define SNP(buf) buf, sizeof(buf)
741 * We enter here when we have a rule with O_LOG.
742 * XXX this function alone takes about 2Kbytes of code!
745 ipfw_log(struct ip_fw *f, u_int hlen, struct ip_fw_args *args,
746 struct mbuf *m, struct ifnet *oif, u_short offset)
748 struct ether_header *eh = args->eh;
750 int limit_reached = 0;
751 char action2[40], proto[128], fragment[32];
756 if (f == NULL) { /* bogus pkt */
757 if (verbose_limit != 0 && norule_counter >= verbose_limit)
760 if (norule_counter == verbose_limit)
761 limit_reached = verbose_limit;
763 } else { /* O_LOG is the first action, find the real one */
764 ipfw_insn *cmd = ACTION_PTR(f);
765 ipfw_insn_log *l = (ipfw_insn_log *)cmd;
767 if (l->max_log != 0 && l->log_left == 0)
770 if (l->log_left == 0)
771 limit_reached = l->max_log;
772 cmd += F_LEN(cmd); /* point to first action */
773 if (cmd->opcode == O_ALTQ) {
774 ipfw_insn_altq *altq = (ipfw_insn_altq *)cmd;
776 snprintf(SNPARGS(action2, 0), "Altq %d",
780 if (cmd->opcode == O_PROB)
784 switch (cmd->opcode) {
790 if (cmd->arg1==ICMP_REJECT_RST)
792 else if (cmd->arg1==ICMP_UNREACH_HOST)
795 snprintf(SNPARGS(action2, 0), "Unreach %d",
800 if (cmd->arg1==ICMP6_UNREACH_RST)
803 snprintf(SNPARGS(action2, 0), "Unreach %d",
814 snprintf(SNPARGS(action2, 0), "Divert %d",
818 snprintf(SNPARGS(action2, 0), "Tee %d",
822 snprintf(SNPARGS(action2, 0), "SkipTo %d",
826 snprintf(SNPARGS(action2, 0), "Pipe %d",
830 snprintf(SNPARGS(action2, 0), "Queue %d",
834 ipfw_insn_sa *sa = (ipfw_insn_sa *)cmd;
837 len = snprintf(SNPARGS(action2, 0), "Forward to %s",
838 inet_ntoa(sa->sa.sin_addr));
840 snprintf(SNPARGS(action2, len), ":%d",
845 snprintf(SNPARGS(action2, 0), "Netgraph %d",
849 snprintf(SNPARGS(action2, 0), "Ngtee %d",
858 if (hlen == 0) { /* non-ip */
859 snprintf(SNPARGS(proto, 0), "MAC");
863 char src[48], dst[48];
864 struct icmphdr *icmp;
867 /* Initialize to make compiler happy. */
868 struct ip *ip = NULL;
870 struct ip6_hdr *ip6 = NULL;
871 struct icmp6_hdr *icmp6;
876 if (args->f_id.addr_type == 6) {
877 snprintf(src, sizeof(src), "[%s]",
878 ip6_sprintf(&args->f_id.src_ip6));
879 snprintf(dst, sizeof(dst), "[%s]",
880 ip6_sprintf(&args->f_id.dst_ip6));
882 ip6 = (struct ip6_hdr *)mtod(m, struct ip6_hdr *);
883 tcp = (struct tcphdr *)(mtod(args->m, char *) + hlen);
884 udp = (struct udphdr *)(mtod(args->m, char *) + hlen);
888 ip = mtod(m, struct ip *);
889 tcp = L3HDR(struct tcphdr, ip);
890 udp = L3HDR(struct udphdr, ip);
892 inet_ntoa_r(ip->ip_src, src);
893 inet_ntoa_r(ip->ip_dst, dst);
896 switch (args->f_id.proto) {
898 len = snprintf(SNPARGS(proto, 0), "TCP %s", src);
900 snprintf(SNPARGS(proto, len), ":%d %s:%d",
901 ntohs(tcp->th_sport),
903 ntohs(tcp->th_dport));
905 snprintf(SNPARGS(proto, len), " %s", dst);
909 len = snprintf(SNPARGS(proto, 0), "UDP %s", src);
911 snprintf(SNPARGS(proto, len), ":%d %s:%d",
912 ntohs(udp->uh_sport),
914 ntohs(udp->uh_dport));
916 snprintf(SNPARGS(proto, len), " %s", dst);
920 icmp = L3HDR(struct icmphdr, ip);
922 len = snprintf(SNPARGS(proto, 0),
924 icmp->icmp_type, icmp->icmp_code);
926 len = snprintf(SNPARGS(proto, 0), "ICMP ");
927 len += snprintf(SNPARGS(proto, len), "%s", src);
928 snprintf(SNPARGS(proto, len), " %s", dst);
932 icmp6 = (struct icmp6_hdr *)(mtod(args->m, char *) + hlen);
934 len = snprintf(SNPARGS(proto, 0),
936 icmp6->icmp6_type, icmp6->icmp6_code);
938 len = snprintf(SNPARGS(proto, 0), "ICMPv6 ");
939 len += snprintf(SNPARGS(proto, len), "%s", src);
940 snprintf(SNPARGS(proto, len), " %s", dst);
944 len = snprintf(SNPARGS(proto, 0), "P:%d %s",
945 args->f_id.proto, src);
946 snprintf(SNPARGS(proto, len), " %s", dst);
951 if (args->f_id.addr_type == 6) {
952 if (offset & (IP6F_OFF_MASK | IP6F_MORE_FRAG))
953 snprintf(SNPARGS(fragment, 0),
954 " (frag %08x:%d@%d%s)",
956 ntohs(ip6->ip6_plen) - hlen,
957 ntohs(offset & IP6F_OFF_MASK) << 3,
958 (offset & IP6F_MORE_FRAG) ? "+" : "");
963 if (eh != NULL) { /* layer 2 packets are as on the wire */
964 ip_off = ntohs(ip->ip_off);
965 ip_len = ntohs(ip->ip_len);
970 if (ip_off & (IP_MF | IP_OFFMASK))
971 snprintf(SNPARGS(fragment, 0),
972 " (frag %d:%d@%d%s)",
973 ntohs(ip->ip_id), ip_len - (ip->ip_hl << 2),
975 (ip_off & IP_MF) ? "+" : "");
978 if (oif || m->m_pkthdr.rcvif)
979 log(LOG_SECURITY | LOG_INFO,
980 "ipfw: %d %s %s %s via %s%s\n",
982 action, proto, oif ? "out" : "in",
983 oif ? oif->if_xname : m->m_pkthdr.rcvif->if_xname,
986 log(LOG_SECURITY | LOG_INFO,
987 "ipfw: %d %s %s [no if info]%s\n",
989 action, proto, fragment);
991 log(LOG_SECURITY | LOG_NOTICE,
992 "ipfw: limit %d reached on entry %d\n",
993 limit_reached, f ? f->rulenum : -1);
997 * IMPORTANT: the hash function for dynamic rules must be commutative
998 * in source and destination (ip,port), because rules are bidirectional
999 * and we want to find both in the same bucket.
1002 hash_packet(struct ipfw_flow_id *id)
1007 if (IS_IP6_FLOW_ID(id))
1008 i = hash_packet6(id);
1011 i = (id->dst_ip) ^ (id->src_ip) ^ (id->dst_port) ^ (id->src_port);
1012 i &= (curr_dyn_buckets - 1);
1017 * unlink a dynamic rule from a chain. prev is a pointer to
1018 * the previous one, q is a pointer to the rule to delete,
1019 * head is a pointer to the head of the queue.
1020 * Modifies q and potentially also head.
1022 #define UNLINK_DYN_RULE(prev, head, q) { \
1023 ipfw_dyn_rule *old_q = q; \
1025 /* remove a refcount to the parent */ \
1026 if (q->dyn_type == O_LIMIT) \
1027 q->parent->count--; \
1028 DEB(printf("ipfw: unlink entry 0x%08x %d -> 0x%08x %d, %d left\n",\
1029 (q->id.src_ip), (q->id.src_port), \
1030 (q->id.dst_ip), (q->id.dst_port), dyn_count-1 ); ) \
1032 prev->next = q = q->next; \
1034 head = q = q->next; \
1036 uma_zfree(ipfw_dyn_rule_zone, old_q); }
1038 #define TIME_LEQ(a,b) ((int)((a)-(b)) <= 0)
1041 * Remove dynamic rules pointing to "rule", or all of them if rule == NULL.
1043 * If keep_me == NULL, rules are deleted even if not expired,
1044 * otherwise only expired rules are removed.
1046 * The value of the second parameter is also used to point to identify
1047 * a rule we absolutely do not want to remove (e.g. because we are
1048 * holding a reference to it -- this is the case with O_LIMIT_PARENT
1049 * rules). The pointer is only used for comparison, so any non-null
1053 remove_dyn_rule(struct ip_fw *rule, ipfw_dyn_rule *keep_me)
1055 static u_int32_t last_remove = 0;
1057 #define FORCE (keep_me == NULL)
1059 ipfw_dyn_rule *prev, *q;
1060 int i, pass = 0, max_pass = 0;
1062 IPFW_DYN_LOCK_ASSERT();
1064 if (ipfw_dyn_v == NULL || dyn_count == 0)
1066 /* do not expire more than once per second, it is useless */
1067 if (!FORCE && last_remove == time_uptime)
1069 last_remove = time_uptime;
1072 * because O_LIMIT refer to parent rules, during the first pass only
1073 * remove child and mark any pending LIMIT_PARENT, and remove
1074 * them in a second pass.
1077 for (i = 0 ; i < curr_dyn_buckets ; i++) {
1078 for (prev=NULL, q = ipfw_dyn_v[i] ; q ; ) {
1080 * Logic can become complex here, so we split tests.
1084 if (rule != NULL && rule != q->rule)
1085 goto next; /* not the one we are looking for */
1086 if (q->dyn_type == O_LIMIT_PARENT) {
1088 * handle parent in the second pass,
1089 * record we need one.
1094 if (FORCE && q->count != 0 ) {
1095 /* XXX should not happen! */
1096 printf("ipfw: OUCH! cannot remove rule,"
1097 " count %d\n", q->count);
1101 !TIME_LEQ( q->expire, time_uptime ))
1104 if (q->dyn_type != O_LIMIT_PARENT || !q->count) {
1105 UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
1113 if (pass++ < max_pass)
1119 * lookup a dynamic rule.
1121 static ipfw_dyn_rule *
1122 lookup_dyn_rule_locked(struct ipfw_flow_id *pkt, int *match_direction,
1126 * stateful ipfw extensions.
1127 * Lookup into dynamic session queue
1129 #define MATCH_REVERSE 0
1130 #define MATCH_FORWARD 1
1131 #define MATCH_NONE 2
1132 #define MATCH_UNKNOWN 3
1133 int i, dir = MATCH_NONE;
1134 ipfw_dyn_rule *prev, *q=NULL;
1136 IPFW_DYN_LOCK_ASSERT();
1138 if (ipfw_dyn_v == NULL)
1139 goto done; /* not found */
1140 i = hash_packet( pkt );
1141 for (prev=NULL, q = ipfw_dyn_v[i] ; q != NULL ; ) {
1142 if (q->dyn_type == O_LIMIT_PARENT && q->count)
1144 if (TIME_LEQ( q->expire, time_uptime)) { /* expire entry */
1145 UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
1148 if (pkt->proto == q->id.proto &&
1149 q->dyn_type != O_LIMIT_PARENT) {
1150 if (IS_IP6_FLOW_ID(pkt)) {
1151 if (IN6_ARE_ADDR_EQUAL(&(pkt->src_ip6),
1152 &(q->id.src_ip6)) &&
1153 IN6_ARE_ADDR_EQUAL(&(pkt->dst_ip6),
1154 &(q->id.dst_ip6)) &&
1155 pkt->src_port == q->id.src_port &&
1156 pkt->dst_port == q->id.dst_port ) {
1157 dir = MATCH_FORWARD;
1160 if (IN6_ARE_ADDR_EQUAL(&(pkt->src_ip6),
1161 &(q->id.dst_ip6)) &&
1162 IN6_ARE_ADDR_EQUAL(&(pkt->dst_ip6),
1163 &(q->id.src_ip6)) &&
1164 pkt->src_port == q->id.dst_port &&
1165 pkt->dst_port == q->id.src_port ) {
1166 dir = MATCH_REVERSE;
1170 if (pkt->src_ip == q->id.src_ip &&
1171 pkt->dst_ip == q->id.dst_ip &&
1172 pkt->src_port == q->id.src_port &&
1173 pkt->dst_port == q->id.dst_port ) {
1174 dir = MATCH_FORWARD;
1177 if (pkt->src_ip == q->id.dst_ip &&
1178 pkt->dst_ip == q->id.src_ip &&
1179 pkt->src_port == q->id.dst_port &&
1180 pkt->dst_port == q->id.src_port ) {
1181 dir = MATCH_REVERSE;
1191 goto done; /* q = NULL, not found */
1193 if ( prev != NULL) { /* found and not in front */
1194 prev->next = q->next;
1195 q->next = ipfw_dyn_v[i];
1198 if (pkt->proto == IPPROTO_TCP) { /* update state according to flags */
1199 u_char flags = pkt->flags & (TH_FIN|TH_SYN|TH_RST);
1201 #define BOTH_SYN (TH_SYN | (TH_SYN << 8))
1202 #define BOTH_FIN (TH_FIN | (TH_FIN << 8))
1203 q->state |= (dir == MATCH_FORWARD ) ? flags : (flags << 8);
1205 case TH_SYN: /* opening */
1206 q->expire = time_uptime + dyn_syn_lifetime;
1209 case BOTH_SYN: /* move to established */
1210 case BOTH_SYN | TH_FIN : /* one side tries to close */
1211 case BOTH_SYN | (TH_FIN << 8) :
1213 #define _SEQ_GE(a,b) ((int)(a) - (int)(b) >= 0)
1214 u_int32_t ack = ntohl(tcp->th_ack);
1215 if (dir == MATCH_FORWARD) {
1216 if (q->ack_fwd == 0 || _SEQ_GE(ack, q->ack_fwd))
1218 else { /* ignore out-of-sequence */
1222 if (q->ack_rev == 0 || _SEQ_GE(ack, q->ack_rev))
1224 else { /* ignore out-of-sequence */
1229 q->expire = time_uptime + dyn_ack_lifetime;
1232 case BOTH_SYN | BOTH_FIN: /* both sides closed */
1233 if (dyn_fin_lifetime >= dyn_keepalive_period)
1234 dyn_fin_lifetime = dyn_keepalive_period - 1;
1235 q->expire = time_uptime + dyn_fin_lifetime;
1241 * reset or some invalid combination, but can also
1242 * occur if we use keep-state the wrong way.
1244 if ( (q->state & ((TH_RST << 8)|TH_RST)) == 0)
1245 printf("invalid state: 0x%x\n", q->state);
1247 if (dyn_rst_lifetime >= dyn_keepalive_period)
1248 dyn_rst_lifetime = dyn_keepalive_period - 1;
1249 q->expire = time_uptime + dyn_rst_lifetime;
1252 } else if (pkt->proto == IPPROTO_UDP) {
1253 q->expire = time_uptime + dyn_udp_lifetime;
1255 /* other protocols */
1256 q->expire = time_uptime + dyn_short_lifetime;
1259 if (match_direction)
1260 *match_direction = dir;
1264 static ipfw_dyn_rule *
1265 lookup_dyn_rule(struct ipfw_flow_id *pkt, int *match_direction,
1271 q = lookup_dyn_rule_locked(pkt, match_direction, tcp);
1274 /* NB: return table locked when q is not NULL */
1279 realloc_dynamic_table(void)
1281 IPFW_DYN_LOCK_ASSERT();
1284 * Try reallocation, make sure we have a power of 2 and do
1285 * not allow more than 64k entries. In case of overflow,
1289 if (dyn_buckets > 65536)
1291 if ((dyn_buckets & (dyn_buckets-1)) != 0) { /* not a power of 2 */
1292 dyn_buckets = curr_dyn_buckets; /* reset */
1295 curr_dyn_buckets = dyn_buckets;
1296 if (ipfw_dyn_v != NULL)
1297 free(ipfw_dyn_v, M_IPFW);
1299 ipfw_dyn_v = malloc(curr_dyn_buckets * sizeof(ipfw_dyn_rule *),
1300 M_IPFW, M_NOWAIT | M_ZERO);
1301 if (ipfw_dyn_v != NULL || curr_dyn_buckets <= 2)
1303 curr_dyn_buckets /= 2;
1308 * Install state of type 'type' for a dynamic session.
1309 * The hash table contains two type of rules:
1310 * - regular rules (O_KEEP_STATE)
1311 * - rules for sessions with limited number of sess per user
1312 * (O_LIMIT). When they are created, the parent is
1313 * increased by 1, and decreased on delete. In this case,
1314 * the third parameter is the parent rule and not the chain.
1315 * - "parent" rules for the above (O_LIMIT_PARENT).
1317 static ipfw_dyn_rule *
1318 add_dyn_rule(struct ipfw_flow_id *id, u_int8_t dyn_type, struct ip_fw *rule)
1323 IPFW_DYN_LOCK_ASSERT();
1325 if (ipfw_dyn_v == NULL ||
1326 (dyn_count == 0 && dyn_buckets != curr_dyn_buckets)) {
1327 realloc_dynamic_table();
1328 if (ipfw_dyn_v == NULL)
1329 return NULL; /* failed ! */
1331 i = hash_packet(id);
1333 r = uma_zalloc(ipfw_dyn_rule_zone, M_NOWAIT | M_ZERO);
1335 printf ("ipfw: sorry cannot allocate state\n");
1339 /* increase refcount on parent, and set pointer */
1340 if (dyn_type == O_LIMIT) {
1341 ipfw_dyn_rule *parent = (ipfw_dyn_rule *)rule;
1342 if ( parent->dyn_type != O_LIMIT_PARENT)
1343 panic("invalid parent");
1346 rule = parent->rule;
1350 r->expire = time_uptime + dyn_syn_lifetime;
1352 r->dyn_type = dyn_type;
1353 r->pcnt = r->bcnt = 0;
1357 r->next = ipfw_dyn_v[i];
1360 DEB(printf("ipfw: add dyn entry ty %d 0x%08x %d -> 0x%08x %d, total %d\n",
1362 (r->id.src_ip), (r->id.src_port),
1363 (r->id.dst_ip), (r->id.dst_port),
1369 * lookup dynamic parent rule using pkt and rule as search keys.
1370 * If the lookup fails, then install one.
1372 static ipfw_dyn_rule *
1373 lookup_dyn_parent(struct ipfw_flow_id *pkt, struct ip_fw *rule)
1378 IPFW_DYN_LOCK_ASSERT();
1381 int is_v6 = IS_IP6_FLOW_ID(pkt);
1382 i = hash_packet( pkt );
1383 for (q = ipfw_dyn_v[i] ; q != NULL ; q=q->next)
1384 if (q->dyn_type == O_LIMIT_PARENT &&
1386 pkt->proto == q->id.proto &&
1387 pkt->src_port == q->id.src_port &&
1388 pkt->dst_port == q->id.dst_port &&
1391 IN6_ARE_ADDR_EQUAL(&(pkt->src_ip6),
1392 &(q->id.src_ip6)) &&
1393 IN6_ARE_ADDR_EQUAL(&(pkt->dst_ip6),
1394 &(q->id.dst_ip6))) ||
1396 pkt->src_ip == q->id.src_ip &&
1397 pkt->dst_ip == q->id.dst_ip)
1400 q->expire = time_uptime + dyn_short_lifetime;
1401 DEB(printf("ipfw: lookup_dyn_parent found 0x%p\n",q);)
1405 return add_dyn_rule(pkt, O_LIMIT_PARENT, rule);
1409 * Install dynamic state for rule type cmd->o.opcode
1411 * Returns 1 (failure) if state is not installed because of errors or because
1412 * session limitations are enforced.
1415 install_state(struct ip_fw *rule, ipfw_insn_limit *cmd,
1416 struct ip_fw_args *args)
1418 static int last_log;
1422 DEB(printf("ipfw: install state type %d 0x%08x %u -> 0x%08x %u\n",
1424 (args->f_id.src_ip), (args->f_id.src_port),
1425 (args->f_id.dst_ip), (args->f_id.dst_port) );)
1429 q = lookup_dyn_rule_locked(&args->f_id, NULL, NULL);
1431 if (q != NULL) { /* should never occur */
1432 if (last_log != time_uptime) {
1433 last_log = time_uptime;
1434 printf("ipfw: install_state: entry already present, done\n");
1440 if (dyn_count >= dyn_max)
1442 * Run out of slots, try to remove any expired rule.
1444 remove_dyn_rule(NULL, (ipfw_dyn_rule *)1);
1446 if (dyn_count >= dyn_max) {
1447 if (last_log != time_uptime) {
1448 last_log = time_uptime;
1449 printf("ipfw: install_state: Too many dynamic rules\n");
1452 return 1; /* cannot install, notify caller */
1455 switch (cmd->o.opcode) {
1456 case O_KEEP_STATE: /* bidir rule */
1457 add_dyn_rule(&args->f_id, O_KEEP_STATE, rule);
1460 case O_LIMIT: /* limit number of sessions */
1462 u_int16_t limit_mask = cmd->limit_mask;
1463 struct ipfw_flow_id id;
1464 ipfw_dyn_rule *parent;
1466 DEB(printf("ipfw: installing dyn-limit rule %d\n",
1469 id.dst_ip = id.src_ip = 0;
1470 id.dst_port = id.src_port = 0;
1471 id.proto = args->f_id.proto;
1473 if (IS_IP6_FLOW_ID (&(args->f_id))) {
1474 if (limit_mask & DYN_SRC_ADDR)
1475 id.src_ip6 = args->f_id.src_ip6;
1476 if (limit_mask & DYN_DST_ADDR)
1477 id.dst_ip6 = args->f_id.dst_ip6;
1479 if (limit_mask & DYN_SRC_ADDR)
1480 id.src_ip = args->f_id.src_ip;
1481 if (limit_mask & DYN_DST_ADDR)
1482 id.dst_ip = args->f_id.dst_ip;
1484 if (limit_mask & DYN_SRC_PORT)
1485 id.src_port = args->f_id.src_port;
1486 if (limit_mask & DYN_DST_PORT)
1487 id.dst_port = args->f_id.dst_port;
1488 parent = lookup_dyn_parent(&id, rule);
1489 if (parent == NULL) {
1490 printf("ipfw: add parent failed\n");
1494 if (parent->count >= cmd->conn_limit) {
1496 * See if we can remove some expired rule.
1498 remove_dyn_rule(rule, parent);
1499 if (parent->count >= cmd->conn_limit) {
1500 if (fw_verbose && last_log != time_uptime) {
1501 last_log = time_uptime;
1502 log(LOG_SECURITY | LOG_DEBUG,
1503 "drop session, too many entries\n");
1509 add_dyn_rule(&args->f_id, O_LIMIT, (struct ip_fw *)parent);
1513 printf("ipfw: unknown dynamic rule type %u\n", cmd->o.opcode);
1517 lookup_dyn_rule_locked(&args->f_id, NULL, NULL); /* XXX just set lifetime */
1523 * Generate a TCP packet, containing either a RST or a keepalive.
1524 * When flags & TH_RST, we are sending a RST packet, because of a
1525 * "reset" action matched the packet.
1526 * Otherwise we are sending a keepalive, and flags & TH_
1528 static struct mbuf *
1529 send_pkt(struct ipfw_flow_id *id, u_int32_t seq, u_int32_t ack, int flags)
1535 MGETHDR(m, M_DONTWAIT, MT_DATA);
1538 m->m_pkthdr.rcvif = (struct ifnet *)0;
1539 m->m_pkthdr.len = m->m_len = sizeof(struct ip) + sizeof(struct tcphdr);
1540 m->m_data += max_linkhdr;
1542 ip = mtod(m, struct ip *);
1543 bzero(ip, m->m_len);
1544 tcp = (struct tcphdr *)(ip + 1); /* no IP options */
1545 ip->ip_p = IPPROTO_TCP;
1548 * Assume we are sending a RST (or a keepalive in the reverse
1549 * direction), swap src and destination addresses and ports.
1551 ip->ip_src.s_addr = htonl(id->dst_ip);
1552 ip->ip_dst.s_addr = htonl(id->src_ip);
1553 tcp->th_sport = htons(id->dst_port);
1554 tcp->th_dport = htons(id->src_port);
1555 if (flags & TH_RST) { /* we are sending a RST */
1556 if (flags & TH_ACK) {
1557 tcp->th_seq = htonl(ack);
1558 tcp->th_ack = htonl(0);
1559 tcp->th_flags = TH_RST;
1563 tcp->th_seq = htonl(0);
1564 tcp->th_ack = htonl(seq);
1565 tcp->th_flags = TH_RST | TH_ACK;
1569 * We are sending a keepalive. flags & TH_SYN determines
1570 * the direction, forward if set, reverse if clear.
1571 * NOTE: seq and ack are always assumed to be correct
1572 * as set by the caller. This may be confusing...
1574 if (flags & TH_SYN) {
1576 * we have to rewrite the correct addresses!
1578 ip->ip_dst.s_addr = htonl(id->dst_ip);
1579 ip->ip_src.s_addr = htonl(id->src_ip);
1580 tcp->th_dport = htons(id->dst_port);
1581 tcp->th_sport = htons(id->src_port);
1583 tcp->th_seq = htonl(seq);
1584 tcp->th_ack = htonl(ack);
1585 tcp->th_flags = TH_ACK;
1588 * set ip_len to the payload size so we can compute
1589 * the tcp checksum on the pseudoheader
1590 * XXX check this, could save a couple of words ?
1592 ip->ip_len = htons(sizeof(struct tcphdr));
1593 tcp->th_sum = in_cksum(m, m->m_pkthdr.len);
1595 * now fill fields left out earlier
1597 ip->ip_ttl = ip_defttl;
1598 ip->ip_len = m->m_pkthdr.len;
1599 m->m_flags |= M_SKIP_FIREWALL;
1604 * sends a reject message, consuming the mbuf passed as an argument.
1607 send_reject(struct ip_fw_args *args, int code, u_short offset, int ip_len)
1610 if (code != ICMP_REJECT_RST) { /* Send an ICMP unreach */
1611 /* We need the IP header in host order for icmp_error(). */
1612 if (args->eh != NULL) {
1613 struct ip *ip = mtod(args->m, struct ip *);
1614 ip->ip_len = ntohs(ip->ip_len);
1615 ip->ip_off = ntohs(ip->ip_off);
1617 icmp_error(args->m, ICMP_UNREACH, code, 0L, 0);
1618 } else if (offset == 0 && args->f_id.proto == IPPROTO_TCP) {
1619 struct tcphdr *const tcp =
1620 L3HDR(struct tcphdr, mtod(args->m, struct ip *));
1621 if ( (tcp->th_flags & TH_RST) == 0) {
1623 m = send_pkt(&(args->f_id), ntohl(tcp->th_seq),
1625 tcp->th_flags | TH_RST);
1627 ip_output(m, NULL, NULL, 0, NULL, NULL);
1637 * Given an ip_fw *, lookup_next_rule will return a pointer
1638 * to the next rule, which can be either the jump
1639 * target (for skipto instructions) or the next one in the list (in
1640 * all other cases including a missing jump target).
1641 * The result is also written in the "next_rule" field of the rule.
1642 * Backward jumps are not allowed, so start looking from the next
1645 * This never returns NULL -- in case we do not have an exact match,
1646 * the next rule is returned. When the ruleset is changed,
1647 * pointers are flushed so we are always correct.
1650 static struct ip_fw *
1651 lookup_next_rule(struct ip_fw *me)
1653 struct ip_fw *rule = NULL;
1656 /* look for action, in case it is a skipto */
1657 cmd = ACTION_PTR(me);
1658 if (cmd->opcode == O_LOG)
1660 if (cmd->opcode == O_ALTQ)
1662 if ( cmd->opcode == O_SKIPTO )
1663 for (rule = me->next; rule ; rule = rule->next)
1664 if (rule->rulenum >= cmd->arg1)
1666 if (rule == NULL) /* failure or not a skipto */
1668 me->next_rule = rule;
1673 add_table_entry(struct ip_fw_chain *ch, uint16_t tbl, in_addr_t addr,
1674 uint8_t mlen, uint32_t value)
1676 struct radix_node_head *rnh;
1677 struct table_entry *ent;
1679 if (tbl >= IPFW_TABLES_MAX)
1681 rnh = ch->tables[tbl];
1682 ent = malloc(sizeof(*ent), M_IPFW_TBL, M_NOWAIT | M_ZERO);
1686 ent->addr.sin_len = ent->mask.sin_len = 8;
1687 ent->mask.sin_addr.s_addr = htonl(mlen ? ~((1 << (32 - mlen)) - 1) : 0);
1688 ent->addr.sin_addr.s_addr = addr & ent->mask.sin_addr.s_addr;
1689 IPFW_WLOCK(&layer3_chain);
1690 if (rnh->rnh_addaddr(&ent->addr, &ent->mask, rnh, (void *)ent) ==
1692 IPFW_WUNLOCK(&layer3_chain);
1693 free(ent, M_IPFW_TBL);
1696 IPFW_WUNLOCK(&layer3_chain);
1701 del_table_entry(struct ip_fw_chain *ch, uint16_t tbl, in_addr_t addr,
1704 struct radix_node_head *rnh;
1705 struct table_entry *ent;
1706 struct sockaddr_in sa, mask;
1708 if (tbl >= IPFW_TABLES_MAX)
1710 rnh = ch->tables[tbl];
1711 sa.sin_len = mask.sin_len = 8;
1712 mask.sin_addr.s_addr = htonl(mlen ? ~((1 << (32 - mlen)) - 1) : 0);
1713 sa.sin_addr.s_addr = addr & mask.sin_addr.s_addr;
1715 ent = (struct table_entry *)rnh->rnh_deladdr(&sa, &mask, rnh);
1721 free(ent, M_IPFW_TBL);
1726 flush_table_entry(struct radix_node *rn, void *arg)
1728 struct radix_node_head * const rnh = arg;
1729 struct table_entry *ent;
1731 ent = (struct table_entry *)
1732 rnh->rnh_deladdr(rn->rn_key, rn->rn_mask, rnh);
1734 free(ent, M_IPFW_TBL);
1739 flush_table(struct ip_fw_chain *ch, uint16_t tbl)
1741 struct radix_node_head *rnh;
1743 IPFW_WLOCK_ASSERT(ch);
1745 if (tbl >= IPFW_TABLES_MAX)
1747 rnh = ch->tables[tbl];
1748 KASSERT(rnh != NULL, ("NULL IPFW table"));
1749 rnh->rnh_walktree(rnh, flush_table_entry, rnh);
1754 flush_tables(struct ip_fw_chain *ch)
1758 IPFW_WLOCK_ASSERT(ch);
1760 for (tbl = 0; tbl < IPFW_TABLES_MAX; tbl++)
1761 flush_table(ch, tbl);
1765 init_tables(struct ip_fw_chain *ch)
1770 for (i = 0; i < IPFW_TABLES_MAX; i++) {
1771 if (!rn_inithead((void **)&ch->tables[i], 32)) {
1772 for (j = 0; j < i; j++) {
1773 (void) flush_table(ch, j);
1782 lookup_table(struct ip_fw_chain *ch, uint16_t tbl, in_addr_t addr,
1785 struct radix_node_head *rnh;
1786 struct table_entry *ent;
1787 struct sockaddr_in sa;
1789 if (tbl >= IPFW_TABLES_MAX)
1791 rnh = ch->tables[tbl];
1793 sa.sin_addr.s_addr = addr;
1794 ent = (struct table_entry *)(rnh->rnh_lookup(&sa, NULL, rnh));
1803 count_table_entry(struct radix_node *rn, void *arg)
1805 u_int32_t * const cnt = arg;
1812 count_table(struct ip_fw_chain *ch, uint32_t tbl, uint32_t *cnt)
1814 struct radix_node_head *rnh;
1816 if (tbl >= IPFW_TABLES_MAX)
1818 rnh = ch->tables[tbl];
1820 rnh->rnh_walktree(rnh, count_table_entry, cnt);
1825 dump_table_entry(struct radix_node *rn, void *arg)
1827 struct table_entry * const n = (struct table_entry *)rn;
1828 ipfw_table * const tbl = arg;
1829 ipfw_table_entry *ent;
1831 if (tbl->cnt == tbl->size)
1833 ent = &tbl->ent[tbl->cnt];
1834 ent->tbl = tbl->tbl;
1835 if (in_nullhost(n->mask.sin_addr))
1838 ent->masklen = 33 - ffs(ntohl(n->mask.sin_addr.s_addr));
1839 ent->addr = n->addr.sin_addr.s_addr;
1840 ent->value = n->value;
1846 dump_table(struct ip_fw_chain *ch, ipfw_table *tbl)
1848 struct radix_node_head *rnh;
1850 if (tbl->tbl >= IPFW_TABLES_MAX)
1852 rnh = ch->tables[tbl->tbl];
1854 rnh->rnh_walktree(rnh, dump_table_entry, tbl);
1859 fill_ugid_cache(struct inpcb *inp, struct ip_fw_ugid *ugp)
1863 if (inp->inp_socket != NULL) {
1864 cr = inp->inp_socket->so_cred;
1865 ugp->fw_prid = jailed(cr) ?
1866 cr->cr_prison->pr_id : -1;
1867 ugp->fw_uid = cr->cr_uid;
1868 ugp->fw_ngroups = cr->cr_ngroups;
1869 bcopy(cr->cr_groups, ugp->fw_groups,
1870 sizeof(ugp->fw_groups));
1875 check_uidgid(ipfw_insn_u32 *insn,
1876 int proto, struct ifnet *oif,
1877 struct in_addr dst_ip, u_int16_t dst_port,
1878 struct in_addr src_ip, u_int16_t src_port,
1879 struct ip_fw_ugid *ugp, int *lookup, struct inpcb *inp)
1881 struct inpcbinfo *pi;
1888 * Check to see if the UDP or TCP stack supplied us with
1889 * the PCB. If so, rather then holding a lock and looking
1890 * up the PCB, we can use the one that was supplied.
1892 if (inp && *lookup == 0) {
1893 INP_LOCK_ASSERT(inp);
1894 if (inp->inp_socket != NULL) {
1895 fill_ugid_cache(inp, ugp);
1900 * If we have already been here and the packet has no
1901 * PCB entry associated with it, then we can safely
1902 * assume that this is a no match.
1906 if (proto == IPPROTO_TCP) {
1909 } else if (proto == IPPROTO_UDP) {
1918 in_pcblookup_hash(pi,
1919 dst_ip, htons(dst_port),
1920 src_ip, htons(src_port),
1922 in_pcblookup_hash(pi,
1923 src_ip, htons(src_port),
1924 dst_ip, htons(dst_port),
1928 if (pcb->inp_socket != NULL) {
1929 fill_ugid_cache(pcb, ugp);
1934 INP_INFO_RUNLOCK(pi);
1937 * If the lookup did not yield any results, there
1938 * is no sense in coming back and trying again. So
1939 * we can set lookup to -1 and ensure that we wont
1940 * bother the pcb system again.
1946 if (insn->o.opcode == O_UID)
1947 match = (ugp->fw_uid == (uid_t)insn->d[0]);
1948 else if (insn->o.opcode == O_GID) {
1949 for (gp = ugp->fw_groups;
1950 gp < &ugp->fw_groups[ugp->fw_ngroups]; gp++)
1951 if (*gp == (gid_t)insn->d[0]) {
1955 } else if (insn->o.opcode == O_JAIL)
1956 match = (ugp->fw_prid == (int)insn->d[0]);
1961 * The main check routine for the firewall.
1963 * All arguments are in args so we can modify them and return them
1964 * back to the caller.
1968 * args->m (in/out) The packet; we set to NULL when/if we nuke it.
1969 * Starts with the IP header.
1970 * args->eh (in) Mac header if present, or NULL for layer3 packet.
1971 * args->oif Outgoing interface, or NULL if packet is incoming.
1972 * The incoming interface is in the mbuf. (in)
1973 * args->divert_rule (in/out)
1974 * Skip up to the first rule past this rule number;
1975 * upon return, non-zero port number for divert or tee.
1977 * args->rule Pointer to the last matching rule (in/out)
1978 * args->next_hop Socket we are forwarding to (out).
1979 * args->f_id Addresses grabbed from the packet (out)
1980 * args->cookie a cookie depending on rule action
1984 * IP_FW_PASS the packet must be accepted
1985 * IP_FW_DENY the packet must be dropped
1986 * IP_FW_DIVERT divert packet, port in m_tag
1987 * IP_FW_TEE tee packet, port in m_tag
1988 * IP_FW_DUMMYNET to dummynet, pipe in args->cookie
1989 * IP_FW_NETGRAPH into netgraph, cookie args->cookie
1994 ipfw_chk(struct ip_fw_args *args)
1997 * Local variables hold state during the processing of a packet.
1999 * IMPORTANT NOTE: to speed up the processing of rules, there
2000 * are some assumption on the values of the variables, which
2001 * are documented here. Should you change them, please check
2002 * the implementation of the various instructions to make sure
2003 * that they still work.
2005 * args->eh The MAC header. It is non-null for a layer2
2006 * packet, it is NULL for a layer-3 packet.
2008 * m | args->m Pointer to the mbuf, as received from the caller.
2009 * It may change if ipfw_chk() does an m_pullup, or if it
2010 * consumes the packet because it calls send_reject().
2011 * XXX This has to change, so that ipfw_chk() never modifies
2012 * or consumes the buffer.
2013 * ip is simply an alias of the value of m, and it is kept
2014 * in sync with it (the packet is supposed to start with
2017 struct mbuf *m = args->m;
2018 struct ip *ip = mtod(m, struct ip *);
2021 * For rules which contain uid/gid or jail constraints, cache
2022 * a copy of the users credentials after the pcb lookup has been
2023 * executed. This will speed up the processing of rules with
2024 * these types of constraints, as well as decrease contention
2025 * on pcb related locks.
2027 struct ip_fw_ugid fw_ugid_cache;
2028 int ugid_lookup = 0;
2031 * divinput_flags If non-zero, set to the IP_FW_DIVERT_*_FLAG
2032 * associated with a packet input on a divert socket. This
2033 * will allow to distinguish traffic and its direction when
2034 * it originates from a divert socket.
2036 u_int divinput_flags = 0;
2039 * oif | args->oif If NULL, ipfw_chk has been called on the
2040 * inbound path (ether_input, ip_input).
2041 * If non-NULL, ipfw_chk has been called on the outbound path
2042 * (ether_output, ip_output).
2044 struct ifnet *oif = args->oif;
2046 struct ip_fw *f = NULL; /* matching rule */
2050 * hlen The length of the IP header.
2052 u_int hlen = 0; /* hlen >0 means we have an IP pkt */
2055 * offset The offset of a fragment. offset != 0 means that
2056 * we have a fragment at this offset of an IPv4 packet.
2057 * offset == 0 means that (if this is an IPv4 packet)
2058 * this is the first or only fragment.
2059 * For IPv6 offset == 0 means there is no Fragment Header.
2060 * If offset != 0 for IPv6 always use correct mask to
2061 * get the correct offset because we add IP6F_MORE_FRAG
2062 * to be able to dectect the first fragment which would
2063 * otherwise have offset = 0.
2068 * Local copies of addresses. They are only valid if we have
2071 * proto The protocol. Set to 0 for non-ip packets,
2072 * or to the protocol read from the packet otherwise.
2073 * proto != 0 means that we have an IPv4 packet.
2075 * src_port, dst_port port numbers, in HOST format. Only
2076 * valid for TCP and UDP packets.
2078 * src_ip, dst_ip ip addresses, in NETWORK format.
2079 * Only valid for IPv4 packets.
2082 u_int16_t src_port = 0, dst_port = 0; /* NOTE: host format */
2083 struct in_addr src_ip, dst_ip; /* NOTE: network format */
2088 * dyn_dir = MATCH_UNKNOWN when rules unchecked,
2089 * MATCH_NONE when checked and not matched (q = NULL),
2090 * MATCH_FORWARD or MATCH_REVERSE otherwise (q != NULL)
2092 int dyn_dir = MATCH_UNKNOWN;
2093 ipfw_dyn_rule *q = NULL;
2094 struct ip_fw_chain *chain = &layer3_chain;
2098 * We store in ulp a pointer to the upper layer protocol header.
2099 * In the ipv4 case this is easy to determine from the header,
2100 * but for ipv6 we might have some additional headers in the middle.
2101 * ulp is NULL if not found.
2103 void *ulp = NULL; /* upper layer protocol pointer. */
2104 /* XXX ipv6 variables */
2106 u_int16_t ext_hd = 0; /* bits vector for extension header filtering */
2107 /* end of ipv6 variables */
2110 if (m->m_flags & M_SKIP_FIREWALL)
2111 return (IP_FW_PASS); /* accept */
2113 pktlen = m->m_pkthdr.len;
2114 proto = args->f_id.proto = 0; /* mark f_id invalid */
2115 /* XXX 0 is a valid proto: IP/IPv6 Hop-by-Hop Option */
2118 * PULLUP_TO(len, p, T) makes sure that len + sizeof(T) is contiguous,
2119 * then it sets p to point at the offset "len" in the mbuf. WARNING: the
2120 * pointer might become stale after other pullups (but we never use it
2123 #define PULLUP_TO(len, p, T) \
2125 int x = (len) + sizeof(T); \
2126 if ((m)->m_len < x) { \
2127 args->m = m = m_pullup(m, x); \
2129 goto pullup_failed; \
2131 p = (mtod(m, char *) + (len)); \
2134 /* Identify IP packets and fill up variables. */
2135 if (pktlen >= sizeof(struct ip6_hdr) &&
2136 (args->eh == NULL || ntohs(args->eh->ether_type)==ETHERTYPE_IPV6) &&
2137 mtod(m, struct ip *)->ip_v == 6) {
2139 args->f_id.addr_type = 6;
2140 hlen = sizeof(struct ip6_hdr);
2141 proto = mtod(m, struct ip6_hdr *)->ip6_nxt;
2143 /* Search extension headers to find upper layer protocols */
2144 while (ulp == NULL) {
2146 case IPPROTO_ICMPV6:
2147 PULLUP_TO(hlen, ulp, struct icmp6_hdr);
2148 args->f_id.flags = ICMP6(ulp)->icmp6_type;
2152 PULLUP_TO(hlen, ulp, struct tcphdr);
2153 dst_port = TCP(ulp)->th_dport;
2154 src_port = TCP(ulp)->th_sport;
2155 args->f_id.flags = TCP(ulp)->th_flags;
2159 PULLUP_TO(hlen, ulp, struct udphdr);
2160 dst_port = UDP(ulp)->uh_dport;
2161 src_port = UDP(ulp)->uh_sport;
2164 case IPPROTO_HOPOPTS: /* RFC 2460 */
2165 PULLUP_TO(hlen, ulp, struct ip6_hbh);
2166 ext_hd |= EXT_HOPOPTS;
2167 hlen += (((struct ip6_hbh *)ulp)->ip6h_len + 1) << 3;
2168 proto = ((struct ip6_hbh *)ulp)->ip6h_nxt;
2172 case IPPROTO_ROUTING: /* RFC 2460 */
2173 PULLUP_TO(hlen, ulp, struct ip6_rthdr);
2174 if (((struct ip6_rthdr *)ulp)->ip6r_type != 0) {
2175 printf("IPFW2: IPV6 - Unknown Routing "
2176 "Header type(%d)\n",
2177 ((struct ip6_rthdr *)ulp)->ip6r_type);
2178 if (fw_deny_unknown_exthdrs)
2179 return (IP_FW_DENY);
2182 ext_hd |= EXT_ROUTING;
2183 hlen += (((struct ip6_rthdr *)ulp)->ip6r_len + 1) << 3;
2184 proto = ((struct ip6_rthdr *)ulp)->ip6r_nxt;
2188 case IPPROTO_FRAGMENT: /* RFC 2460 */
2189 PULLUP_TO(hlen, ulp, struct ip6_frag);
2190 ext_hd |= EXT_FRAGMENT;
2191 hlen += sizeof (struct ip6_frag);
2192 proto = ((struct ip6_frag *)ulp)->ip6f_nxt;
2193 offset = ((struct ip6_frag *)ulp)->ip6f_offlg &
2195 /* Add IP6F_MORE_FRAG for offset of first
2196 * fragment to be != 0. */
2197 offset |= ((struct ip6_frag *)ulp)->ip6f_offlg &
2200 printf("IPFW2: IPV6 - Invalid Fragment "
2202 if (fw_deny_unknown_exthdrs)
2203 return (IP_FW_DENY);
2206 args->f_id.frag_id6 =
2207 ntohl(((struct ip6_frag *)ulp)->ip6f_ident);
2211 case IPPROTO_DSTOPTS: /* RFC 2460 */
2212 PULLUP_TO(hlen, ulp, struct ip6_hbh);
2213 ext_hd |= EXT_DSTOPTS;
2214 hlen += (((struct ip6_hbh *)ulp)->ip6h_len + 1) << 3;
2215 proto = ((struct ip6_hbh *)ulp)->ip6h_nxt;
2219 case IPPROTO_AH: /* RFC 2402 */
2220 PULLUP_TO(hlen, ulp, struct ip6_ext);
2222 hlen += (((struct ip6_ext *)ulp)->ip6e_len + 2) << 2;
2223 proto = ((struct ip6_ext *)ulp)->ip6e_nxt;
2227 case IPPROTO_ESP: /* RFC 2406 */
2228 PULLUP_TO(hlen, ulp, uint32_t); /* SPI, Seq# */
2229 /* Anything past Seq# is variable length and
2230 * data past this ext. header is encrypted. */
2234 case IPPROTO_NONE: /* RFC 2460 */
2235 PULLUP_TO(hlen, ulp, struct ip6_ext);
2236 /* Packet ends here. if ip6e_len!=0 octets
2237 * must be ignored. */
2240 case IPPROTO_OSPFIGP:
2241 /* XXX OSPF header check? */
2242 PULLUP_TO(hlen, ulp, struct ip6_ext);
2246 printf("IPFW2: IPV6 - Unknown Extension "
2247 "Header(%d), ext_hd=%x\n", proto, ext_hd);
2248 if (fw_deny_unknown_exthdrs)
2249 return (IP_FW_DENY);
2253 args->f_id.src_ip6 = mtod(m,struct ip6_hdr *)->ip6_src;
2254 args->f_id.dst_ip6 = mtod(m,struct ip6_hdr *)->ip6_dst;
2255 args->f_id.src_ip = 0;
2256 args->f_id.dst_ip = 0;
2257 args->f_id.flow_id6 = ntohl(mtod(m, struct ip6_hdr *)->ip6_flow);
2258 } else if (pktlen >= sizeof(struct ip) &&
2259 (args->eh == NULL || ntohs(args->eh->ether_type) == ETHERTYPE_IP) &&
2260 mtod(m, struct ip *)->ip_v == 4) {
2262 ip = mtod(m, struct ip *);
2263 hlen = ip->ip_hl << 2;
2264 args->f_id.addr_type = 4;
2267 * Collect parameters into local variables for faster matching.
2270 src_ip = ip->ip_src;
2271 dst_ip = ip->ip_dst;
2272 if (args->eh != NULL) { /* layer 2 packets are as on the wire */
2273 offset = ntohs(ip->ip_off) & IP_OFFMASK;
2274 ip_len = ntohs(ip->ip_len);
2276 offset = ip->ip_off & IP_OFFMASK;
2277 ip_len = ip->ip_len;
2279 pktlen = ip_len < pktlen ? ip_len : pktlen;
2284 PULLUP_TO(hlen, ulp, struct tcphdr);
2285 dst_port = TCP(ulp)->th_dport;
2286 src_port = TCP(ulp)->th_sport;
2287 args->f_id.flags = TCP(ulp)->th_flags;
2291 PULLUP_TO(hlen, ulp, struct udphdr);
2292 dst_port = UDP(ulp)->uh_dport;
2293 src_port = UDP(ulp)->uh_sport;
2297 PULLUP_TO(hlen, ulp, struct icmphdr);
2298 args->f_id.flags = ICMP(ulp)->icmp_type;
2306 args->f_id.src_ip = ntohl(src_ip.s_addr);
2307 args->f_id.dst_ip = ntohl(dst_ip.s_addr);
2310 if (proto) { /* we may have port numbers, store them */
2311 args->f_id.proto = proto;
2312 args->f_id.src_port = src_port = ntohs(src_port);
2313 args->f_id.dst_port = dst_port = ntohs(dst_port);
2317 mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL);
2320 * Packet has already been tagged. Look for the next rule
2321 * to restart processing.
2323 * If fw_one_pass != 0 then just accept it.
2324 * XXX should not happen here, but optimized out in
2328 IPFW_RUNLOCK(chain);
2329 return (IP_FW_PASS);
2332 f = args->rule->next_rule;
2334 f = lookup_next_rule(args->rule);
2337 * Find the starting rule. It can be either the first
2338 * one, or the one after divert_rule if asked so.
2340 int skipto = mtag ? divert_cookie(mtag) : 0;
2343 if (args->eh == NULL && skipto != 0) {
2344 if (skipto >= IPFW_DEFAULT_RULE) {
2345 IPFW_RUNLOCK(chain);
2346 return (IP_FW_DENY); /* invalid */
2348 while (f && f->rulenum <= skipto)
2350 if (f == NULL) { /* drop packet */
2351 IPFW_RUNLOCK(chain);
2352 return (IP_FW_DENY);
2356 /* reset divert rule to avoid confusion later */
2358 divinput_flags = divert_info(mtag) &
2359 (IP_FW_DIVERT_OUTPUT_FLAG | IP_FW_DIVERT_LOOPBACK_FLAG);
2360 m_tag_delete(m, mtag);
2364 * Now scan the rules, and parse microinstructions for each rule.
2366 for (; f; f = f->next) {
2368 uint32_t tablearg = 0;
2369 int l, cmdlen, skip_or; /* skip rest of OR block */
2372 if (set_disable & (1 << f->set) )
2376 for (l = f->cmd_len, cmd = f->cmd ; l > 0 ;
2377 l -= cmdlen, cmd += cmdlen) {
2381 * check_body is a jump target used when we find a
2382 * CHECK_STATE, and need to jump to the body of
2387 cmdlen = F_LEN(cmd);
2389 * An OR block (insn_1 || .. || insn_n) has the
2390 * F_OR bit set in all but the last instruction.
2391 * The first match will set "skip_or", and cause
2392 * the following instructions to be skipped until
2393 * past the one with the F_OR bit clear.
2395 if (skip_or) { /* skip this instruction */
2396 if ((cmd->len & F_OR) == 0)
2397 skip_or = 0; /* next one is good */
2400 match = 0; /* set to 1 if we succeed */
2402 switch (cmd->opcode) {
2404 * The first set of opcodes compares the packet's
2405 * fields with some pattern, setting 'match' if a
2406 * match is found. At the end of the loop there is
2407 * logic to deal with F_NOT and F_OR flags associated
2415 printf("ipfw: opcode %d unimplemented\n",
2423 * We only check offset == 0 && proto != 0,
2424 * as this ensures that we have a
2425 * packet with the ports info.
2429 if (is_ipv6) /* XXX to be fixed later */
2431 if (proto == IPPROTO_TCP ||
2432 proto == IPPROTO_UDP)
2433 match = check_uidgid(
2434 (ipfw_insn_u32 *)cmd,
2437 src_ip, src_port, &fw_ugid_cache,
2438 &ugid_lookup, args->inp);
2442 match = iface_match(m->m_pkthdr.rcvif,
2443 (ipfw_insn_if *)cmd);
2447 match = iface_match(oif, (ipfw_insn_if *)cmd);
2451 match = iface_match(oif ? oif :
2452 m->m_pkthdr.rcvif, (ipfw_insn_if *)cmd);
2456 if (args->eh != NULL) { /* have MAC header */
2457 u_int32_t *want = (u_int32_t *)
2458 ((ipfw_insn_mac *)cmd)->addr;
2459 u_int32_t *mask = (u_int32_t *)
2460 ((ipfw_insn_mac *)cmd)->mask;
2461 u_int32_t *hdr = (u_int32_t *)args->eh;
2464 ( want[0] == (hdr[0] & mask[0]) &&
2465 want[1] == (hdr[1] & mask[1]) &&
2466 want[2] == (hdr[2] & mask[2]) );
2471 if (args->eh != NULL) {
2473 ntohs(args->eh->ether_type);
2475 ((ipfw_insn_u16 *)cmd)->ports;
2478 for (i = cmdlen - 1; !match && i>0;
2480 match = (t>=p[0] && t<=p[1]);
2485 match = (offset != 0);
2488 case O_IN: /* "out" is "not in" */
2489 match = (oif == NULL);
2493 match = (args->eh != NULL);
2497 match = (cmd->arg1 & 1 && divinput_flags &
2498 IP_FW_DIVERT_LOOPBACK_FLAG) ||
2499 (cmd->arg1 & 2 && divinput_flags &
2500 IP_FW_DIVERT_OUTPUT_FLAG);
2505 * We do not allow an arg of 0 so the
2506 * check of "proto" only suffices.
2508 match = (proto == cmd->arg1);
2513 (((ipfw_insn_ip *)cmd)->addr.s_addr ==
2517 case O_IP_SRC_LOOKUP:
2518 case O_IP_DST_LOOKUP:
2521 (cmd->opcode == O_IP_DST_LOOKUP) ?
2522 dst_ip.s_addr : src_ip.s_addr;
2525 match = lookup_table(chain, cmd->arg1, a,
2529 if (cmdlen == F_INSN_SIZE(ipfw_insn_u32))
2531 ((ipfw_insn_u32 *)cmd)->d[0] == v;
2541 (cmd->opcode == O_IP_DST_MASK) ?
2542 dst_ip.s_addr : src_ip.s_addr;
2543 uint32_t *p = ((ipfw_insn_u32 *)cmd)->d;
2546 for (; !match && i>0; i-= 2, p+= 2)
2547 match = (p[0] == (a & p[1]));
2555 INADDR_TO_IFP(src_ip, tif);
2556 match = (tif != NULL);
2563 u_int32_t *d = (u_int32_t *)(cmd+1);
2565 cmd->opcode == O_IP_DST_SET ?
2571 addr -= d[0]; /* subtract base */
2572 match = (addr < cmd->arg1) &&
2573 ( d[ 1 + (addr>>5)] &
2574 (1<<(addr & 0x1f)) );
2580 (((ipfw_insn_ip *)cmd)->addr.s_addr ==
2588 INADDR_TO_IFP(dst_ip, tif);
2589 match = (tif != NULL);
2596 * offset == 0 && proto != 0 is enough
2597 * to guarantee that we have a
2598 * packet with port info.
2600 if ((proto==IPPROTO_UDP || proto==IPPROTO_TCP)
2603 (cmd->opcode == O_IP_SRCPORT) ?
2604 src_port : dst_port ;
2606 ((ipfw_insn_u16 *)cmd)->ports;
2609 for (i = cmdlen - 1; !match && i>0;
2611 match = (x>=p[0] && x<=p[1]);
2616 match = (offset == 0 && proto==IPPROTO_ICMP &&
2617 icmptype_match(ICMP(ulp), (ipfw_insn_u32 *)cmd) );
2622 match = is_ipv6 && offset == 0 &&
2623 proto==IPPROTO_ICMPV6 &&
2625 ICMP6(ulp)->icmp6_type,
2626 (ipfw_insn_u32 *)cmd);
2632 ipopts_match(mtod(m, struct ip *), cmd) );
2637 cmd->arg1 == mtod(m, struct ip *)->ip_v);
2643 if (is_ipv4) { /* only for IP packets */
2648 if (cmd->opcode == O_IPLEN)
2650 else if (cmd->opcode == O_IPTTL)
2651 x = mtod(m, struct ip *)->ip_ttl;
2652 else /* must be IPID */
2653 x = ntohs(mtod(m, struct ip *)->ip_id);
2655 match = (cmd->arg1 == x);
2658 /* otherwise we have ranges */
2659 p = ((ipfw_insn_u16 *)cmd)->ports;
2661 for (; !match && i>0; i--, p += 2)
2662 match = (x >= p[0] && x <= p[1]);
2666 case O_IPPRECEDENCE:
2668 (cmd->arg1 == (mtod(m, struct ip *)->ip_tos & 0xe0)) );
2673 flags_match(cmd, mtod(m, struct ip *)->ip_tos));
2677 if (proto == IPPROTO_TCP && offset == 0) {
2685 ((ip->ip_hl + tcp->th_off) << 2);
2687 match = (cmd->arg1 == x);
2690 /* otherwise we have ranges */
2691 p = ((ipfw_insn_u16 *)cmd)->ports;
2693 for (; !match && i>0; i--, p += 2)
2694 match = (x >= p[0] && x <= p[1]);
2699 match = (proto == IPPROTO_TCP && offset == 0 &&
2700 flags_match(cmd, TCP(ulp)->th_flags));
2704 match = (proto == IPPROTO_TCP && offset == 0 &&
2705 tcpopts_match(TCP(ulp), cmd));
2709 match = (proto == IPPROTO_TCP && offset == 0 &&
2710 ((ipfw_insn_u32 *)cmd)->d[0] ==
2715 match = (proto == IPPROTO_TCP && offset == 0 &&
2716 ((ipfw_insn_u32 *)cmd)->d[0] ==
2721 match = (proto == IPPROTO_TCP && offset == 0 &&
2722 cmd->arg1 == TCP(ulp)->th_win);
2726 /* reject packets which have SYN only */
2727 /* XXX should i also check for TH_ACK ? */
2728 match = (proto == IPPROTO_TCP && offset == 0 &&
2729 (TCP(ulp)->th_flags &
2730 (TH_RST | TH_ACK | TH_SYN)) != TH_SYN);
2734 struct altq_tag *at;
2735 ipfw_insn_altq *altq = (ipfw_insn_altq *)cmd;
2738 mtag = m_tag_find(m, PACKET_TAG_PF_QID, NULL);
2741 mtag = m_tag_get(PACKET_TAG_PF_QID,
2742 sizeof(struct altq_tag),
2746 * Let the packet fall back to the
2751 at = (struct altq_tag *)(mtag+1);
2752 at->qid = altq->qid;
2758 m_tag_prepend(m, mtag);
2764 ipfw_log(f, hlen, args, m, oif, offset);
2769 match = (random()<((ipfw_insn_u32 *)cmd)->d[0]);
2773 /* Outgoing packets automatically pass/match */
2774 match = ((oif != NULL) ||
2775 (m->m_pkthdr.rcvif == NULL) ||
2779 verify_path6(&(args->f_id.src_ip6),
2780 m->m_pkthdr.rcvif) :
2782 verify_path(src_ip, m->m_pkthdr.rcvif)));
2786 /* Outgoing packets automatically pass/match */
2787 match = (hlen > 0 && ((oif != NULL) ||
2790 verify_path6(&(args->f_id.src_ip6),
2793 verify_path(src_ip, NULL)));
2797 /* Outgoing packets automatically pass/match */
2798 if (oif == NULL && hlen > 0 &&
2799 ( (is_ipv4 && in_localaddr(src_ip))
2802 in6_localaddr(&(args->f_id.src_ip6)))
2807 is_ipv6 ? verify_path6(
2808 &(args->f_id.src_ip6),
2809 m->m_pkthdr.rcvif) :
2819 match = (m_tag_find(m,
2820 PACKET_TAG_IPSEC_IN_DONE, NULL) != NULL);
2823 match = (ipsec_getnhist(m) != 0);
2825 /* otherwise no match */
2831 IN6_ARE_ADDR_EQUAL(&args->f_id.src_ip6,
2832 &((ipfw_insn_ip6 *)cmd)->addr6);
2837 IN6_ARE_ADDR_EQUAL(&args->f_id.dst_ip6,
2838 &((ipfw_insn_ip6 *)cmd)->addr6);
2840 case O_IP6_SRC_MASK:
2842 ipfw_insn_ip6 *te = (ipfw_insn_ip6 *)cmd;
2843 struct in6_addr p = args->f_id.src_ip6;
2845 APPLY_MASK(&p, &te->mask6);
2846 match = IN6_ARE_ADDR_EQUAL(&te->addr6, &p);
2850 case O_IP6_DST_MASK:
2852 ipfw_insn_ip6 *te = (ipfw_insn_ip6 *)cmd;
2853 struct in6_addr p = args->f_id.dst_ip6;
2855 APPLY_MASK(&p, &te->mask6);
2856 match = IN6_ARE_ADDR_EQUAL(&te->addr6, &p);
2861 match= is_ipv6 && search_ip6_addr_net(&args->f_id.src_ip6);
2865 match= is_ipv6 && search_ip6_addr_net(&args->f_id.dst_ip6);
2870 flow6id_match(args->f_id.flow_id6,
2871 (ipfw_insn_u32 *) cmd);
2876 (ext_hd & ((ipfw_insn *) cmd)->arg1);
2889 * The second set of opcodes represents 'actions',
2890 * i.e. the terminal part of a rule once the packet
2891 * matches all previous patterns.
2892 * Typically there is only one action for each rule,
2893 * and the opcode is stored at the end of the rule
2894 * (but there are exceptions -- see below).
2896 * In general, here we set retval and terminate the
2897 * outer loop (would be a 'break 3' in some language,
2898 * but we need to do a 'goto done').
2901 * O_COUNT and O_SKIPTO actions:
2902 * instead of terminating, we jump to the next rule
2903 * ('goto next_rule', equivalent to a 'break 2'),
2904 * or to the SKIPTO target ('goto again' after
2905 * having set f, cmd and l), respectively.
2907 * O_LOG and O_ALTQ action parameters:
2908 * perform some action and set match = 1;
2910 * O_LIMIT and O_KEEP_STATE: these opcodes are
2911 * not real 'actions', and are stored right
2912 * before the 'action' part of the rule.
2913 * These opcodes try to install an entry in the
2914 * state tables; if successful, we continue with
2915 * the next opcode (match=1; break;), otherwise
2916 * the packet * must be dropped
2917 * ('goto done' after setting retval);
2919 * O_PROBE_STATE and O_CHECK_STATE: these opcodes
2920 * cause a lookup of the state table, and a jump
2921 * to the 'action' part of the parent rule
2922 * ('goto check_body') if an entry is found, or
2923 * (CHECK_STATE only) a jump to the next rule if
2924 * the entry is not found ('goto next_rule').
2925 * The result of the lookup is cached to make
2926 * further instances of these opcodes are
2931 if (install_state(f,
2932 (ipfw_insn_limit *)cmd, args)) {
2933 retval = IP_FW_DENY;
2934 goto done; /* error/limit violation */
2942 * dynamic rules are checked at the first
2943 * keep-state or check-state occurrence,
2944 * with the result being stored in dyn_dir.
2945 * The compiler introduces a PROBE_STATE
2946 * instruction for us when we have a
2947 * KEEP_STATE (because PROBE_STATE needs
2950 if (dyn_dir == MATCH_UNKNOWN &&
2951 (q = lookup_dyn_rule(&args->f_id,
2952 &dyn_dir, proto == IPPROTO_TCP ?
2956 * Found dynamic entry, update stats
2957 * and jump to the 'action' part of
2963 cmd = ACTION_PTR(f);
2964 l = f->cmd_len - f->act_ofs;
2969 * Dynamic entry not found. If CHECK_STATE,
2970 * skip to next rule, if PROBE_STATE just
2971 * ignore and continue with next opcode.
2973 if (cmd->opcode == O_CHECK_STATE)
2979 retval = 0; /* accept */
2984 args->rule = f; /* report matching rule */
2985 if (cmd->arg1 == IP_FW_TABLEARG)
2986 args->cookie = tablearg;
2988 args->cookie = cmd->arg1;
2989 retval = IP_FW_DUMMYNET;
2994 struct divert_tag *dt;
2996 if (args->eh) /* not on layer 2 */
2998 mtag = m_tag_get(PACKET_TAG_DIVERT,
2999 sizeof(struct divert_tag),
3004 IPFW_RUNLOCK(chain);
3005 return (IP_FW_DENY);
3007 dt = (struct divert_tag *)(mtag+1);
3008 dt->cookie = f->rulenum;
3009 if (cmd->arg1 == IP_FW_TABLEARG)
3010 dt->info = tablearg;
3012 dt->info = cmd->arg1;
3013 m_tag_prepend(m, mtag);
3014 retval = (cmd->opcode == O_DIVERT) ?
3015 IP_FW_DIVERT : IP_FW_TEE;
3021 f->pcnt++; /* update stats */
3023 f->timestamp = time_uptime;
3024 if (cmd->opcode == O_COUNT)
3027 if (f->next_rule == NULL)
3028 lookup_next_rule(f);
3034 * Drop the packet and send a reject notice
3035 * if the packet is not ICMP (or is an ICMP
3036 * query), and it is not multicast/broadcast.
3038 if (hlen > 0 && is_ipv4 && offset == 0 &&
3039 (proto != IPPROTO_ICMP ||
3040 is_icmp_query(ICMP(ulp))) &&
3041 !(m->m_flags & (M_BCAST|M_MCAST)) &&
3042 !IN_MULTICAST(ntohl(dst_ip.s_addr))) {
3043 send_reject(args, cmd->arg1,
3050 if (hlen > 0 && is_ipv6 &&
3051 (proto != IPPROTO_ICMPV6 ||
3052 (is_icmp6_query(args->f_id.flags) == 1)) &&
3053 !(m->m_flags & (M_BCAST|M_MCAST)) &&
3054 !IN6_IS_ADDR_MULTICAST(&args->f_id.dst_ip6)) {
3055 send_reject6(args, cmd->arg1,
3062 retval = IP_FW_DENY;
3066 if (args->eh) /* not valid on layer2 pkts */
3068 if (!q || dyn_dir == MATCH_FORWARD)
3070 &((ipfw_insn_sa *)cmd)->sa;
3071 retval = IP_FW_PASS;
3076 args->rule = f; /* report matching rule */
3077 if (cmd->arg1 == IP_FW_TABLEARG)
3078 args->cookie = tablearg;
3080 args->cookie = cmd->arg1;
3081 retval = (cmd->opcode == O_NETGRAPH) ?
3082 IP_FW_NETGRAPH : IP_FW_NGTEE;
3086 panic("-- unknown opcode %d\n", cmd->opcode);
3087 } /* end of switch() on opcodes */
3089 if (cmd->len & F_NOT)
3093 if (cmd->len & F_OR)
3096 if (!(cmd->len & F_OR)) /* not an OR block, */
3097 break; /* try next rule */
3100 } /* end of inner for, scan opcodes */
3102 next_rule:; /* try next rule */
3104 } /* end of outer for, scan rules */
3105 printf("ipfw: ouch!, skip past end of rules, denying packet\n");
3106 IPFW_RUNLOCK(chain);
3107 return (IP_FW_DENY);
3110 /* Update statistics */
3113 f->timestamp = time_uptime;
3114 IPFW_RUNLOCK(chain);
3119 printf("ipfw: pullup failed\n");
3120 return (IP_FW_DENY);
3124 * When a rule is added/deleted, clear the next_rule pointers in all rules.
3125 * These will be reconstructed on the fly as packets are matched.
3128 flush_rule_ptrs(struct ip_fw_chain *chain)
3132 IPFW_WLOCK_ASSERT(chain);
3134 for (rule = chain->rules; rule; rule = rule->next)
3135 rule->next_rule = NULL;
3139 * Add a new rule to the list. Copy the rule into a malloc'ed area, then
3140 * possibly create a rule number and add the rule to the list.
3141 * Update the rule_number in the input struct so the caller knows it as well.
3144 add_rule(struct ip_fw_chain *chain, struct ip_fw *input_rule)
3146 struct ip_fw *rule, *f, *prev;
3147 int l = RULESIZE(input_rule);
3149 if (chain->rules == NULL && input_rule->rulenum != IPFW_DEFAULT_RULE)
3152 rule = malloc(l, M_IPFW, M_NOWAIT | M_ZERO);
3156 bcopy(input_rule, rule, l);
3159 rule->next_rule = NULL;
3163 rule->timestamp = 0;
3167 if (chain->rules == NULL) { /* default rule */
3168 chain->rules = rule;
3173 * If rulenum is 0, find highest numbered rule before the
3174 * default rule, and add autoinc_step
3176 if (autoinc_step < 1)
3178 else if (autoinc_step > 1000)
3179 autoinc_step = 1000;
3180 if (rule->rulenum == 0) {
3182 * locate the highest numbered rule before default
3184 for (f = chain->rules; f; f = f->next) {
3185 if (f->rulenum == IPFW_DEFAULT_RULE)
3187 rule->rulenum = f->rulenum;
3189 if (rule->rulenum < IPFW_DEFAULT_RULE - autoinc_step)
3190 rule->rulenum += autoinc_step;
3191 input_rule->rulenum = rule->rulenum;
3195 * Now insert the new rule in the right place in the sorted list.
3197 for (prev = NULL, f = chain->rules; f; prev = f, f = f->next) {
3198 if (f->rulenum > rule->rulenum) { /* found the location */
3202 } else { /* head insert */
3203 rule->next = chain->rules;
3204 chain->rules = rule;
3209 flush_rule_ptrs(chain);
3213 IPFW_WUNLOCK(chain);
3214 DEB(printf("ipfw: installed rule %d, static count now %d\n",
3215 rule->rulenum, static_count);)
3220 * Remove a static rule (including derived * dynamic rules)
3221 * and place it on the ``reap list'' for later reclamation.
3222 * The caller is in charge of clearing rule pointers to avoid
3223 * dangling pointers.
3224 * @return a pointer to the next entry.
3225 * Arguments are not checked, so they better be correct.
3227 static struct ip_fw *
3228 remove_rule(struct ip_fw_chain *chain, struct ip_fw *rule, struct ip_fw *prev)
3231 int l = RULESIZE(rule);
3233 IPFW_WLOCK_ASSERT(chain);
3237 remove_dyn_rule(rule, NULL /* force removal */);
3246 rule->next = chain->reap;
3253 * Reclaim storage associated with a list of rules. This is
3254 * typically the list created using remove_rule.
3257 reap_rules(struct ip_fw *head)
3261 while ((rule = head) != NULL) {
3263 if (DUMMYNET_LOADED)
3264 ip_dn_ruledel_ptr(rule);
3270 * Remove all rules from a chain (except rules in set RESVD_SET
3271 * unless kill_default = 1). The caller is responsible for
3272 * reclaiming storage for the rules left in chain->reap.
3275 free_chain(struct ip_fw_chain *chain, int kill_default)
3277 struct ip_fw *prev, *rule;
3279 IPFW_WLOCK_ASSERT(chain);
3281 flush_rule_ptrs(chain); /* more efficient to do outside the loop */
3282 for (prev = NULL, rule = chain->rules; rule ; )
3283 if (kill_default || rule->set != RESVD_SET)
3284 rule = remove_rule(chain, rule, prev);
3292 * Remove all rules with given number, and also do set manipulation.
3293 * Assumes chain != NULL && *chain != NULL.
3295 * The argument is an u_int32_t. The low 16 bit are the rule or set number,
3296 * the next 8 bits are the new set, the top 8 bits are the command:
3298 * 0 delete rules with given number
3299 * 1 delete rules with given set number
3300 * 2 move rules with given number to new set
3301 * 3 move rules with given set number to new set
3302 * 4 swap sets with given numbers
3305 del_entry(struct ip_fw_chain *chain, u_int32_t arg)
3307 struct ip_fw *prev = NULL, *rule;
3308 u_int16_t rulenum; /* rule or old_set */
3309 u_int8_t cmd, new_set;
3311 rulenum = arg & 0xffff;
3312 cmd = (arg >> 24) & 0xff;
3313 new_set = (arg >> 16) & 0xff;
3317 if (new_set > RESVD_SET)
3319 if (cmd == 0 || cmd == 2) {
3320 if (rulenum >= IPFW_DEFAULT_RULE)
3323 if (rulenum > RESVD_SET) /* old_set */
3328 rule = chain->rules;
3331 case 0: /* delete rules with given number */
3333 * locate first rule to delete
3335 for (; rule->rulenum < rulenum; prev = rule, rule = rule->next)
3337 if (rule->rulenum != rulenum) {
3338 IPFW_WUNLOCK(chain);
3343 * flush pointers outside the loop, then delete all matching
3344 * rules. prev remains the same throughout the cycle.
3346 flush_rule_ptrs(chain);
3347 while (rule->rulenum == rulenum)
3348 rule = remove_rule(chain, rule, prev);
3351 case 1: /* delete all rules with given set number */
3352 flush_rule_ptrs(chain);
3353 rule = chain->rules;
3354 while (rule->rulenum < IPFW_DEFAULT_RULE)
3355 if (rule->set == rulenum)
3356 rule = remove_rule(chain, rule, prev);
3363 case 2: /* move rules with given number to new set */
3364 rule = chain->rules;
3365 for (; rule->rulenum < IPFW_DEFAULT_RULE; rule = rule->next)
3366 if (rule->rulenum == rulenum)
3367 rule->set = new_set;
3370 case 3: /* move rules with given set number to new set */
3371 for (; rule->rulenum < IPFW_DEFAULT_RULE; rule = rule->next)
3372 if (rule->set == rulenum)
3373 rule->set = new_set;
3376 case 4: /* swap two sets */
3377 for (; rule->rulenum < IPFW_DEFAULT_RULE; rule = rule->next)
3378 if (rule->set == rulenum)
3379 rule->set = new_set;
3380 else if (rule->set == new_set)
3381 rule->set = rulenum;
3385 * Look for rules to reclaim. We grab the list before
3386 * releasing the lock then reclaim them w/o the lock to
3387 * avoid a LOR with dummynet.
3391 IPFW_WUNLOCK(chain);
3398 * Clear counters for a specific rule.
3399 * The enclosing "table" is assumed locked.
3402 clear_counters(struct ip_fw *rule, int log_only)
3404 ipfw_insn_log *l = (ipfw_insn_log *)ACTION_PTR(rule);
3406 if (log_only == 0) {
3407 rule->bcnt = rule->pcnt = 0;
3408 rule->timestamp = 0;
3410 if (l->o.opcode == O_LOG)
3411 l->log_left = l->max_log;
3415 * Reset some or all counters on firewall rules.
3416 * @arg frwl is null to clear all entries, or contains a specific
3418 * @arg log_only is 1 if we only want to reset logs, zero otherwise.
3421 zero_entry(struct ip_fw_chain *chain, int rulenum, int log_only)
3429 for (rule = chain->rules; rule; rule = rule->next)
3430 clear_counters(rule, log_only);
3431 msg = log_only ? "ipfw: All logging counts reset.\n" :
3432 "ipfw: Accounting cleared.\n";
3436 * We can have multiple rules with the same number, so we
3437 * need to clear them all.
3439 for (rule = chain->rules; rule; rule = rule->next)
3440 if (rule->rulenum == rulenum) {
3441 while (rule && rule->rulenum == rulenum) {
3442 clear_counters(rule, log_only);
3448 if (!cleared) { /* we did not find any matching rules */
3449 IPFW_WUNLOCK(chain);
3452 msg = log_only ? "ipfw: Entry %d logging count reset.\n" :
3453 "ipfw: Entry %d cleared.\n";
3455 IPFW_WUNLOCK(chain);
3458 log(LOG_SECURITY | LOG_NOTICE, msg, rulenum);
3463 * Check validity of the structure before insert.
3464 * Fortunately rules are simple, so this mostly need to check rule sizes.
3467 check_ipfw_struct(struct ip_fw *rule, int size)
3473 if (size < sizeof(*rule)) {
3474 printf("ipfw: rule too short\n");
3477 /* first, check for valid size */
3480 printf("ipfw: size mismatch (have %d want %d)\n", size, l);
3483 if (rule->act_ofs >= rule->cmd_len) {
3484 printf("ipfw: bogus action offset (%u > %u)\n",
3485 rule->act_ofs, rule->cmd_len - 1);
3489 * Now go for the individual checks. Very simple ones, basically only
3490 * instruction sizes.
3492 for (l = rule->cmd_len, cmd = rule->cmd ;
3493 l > 0 ; l -= cmdlen, cmd += cmdlen) {
3494 cmdlen = F_LEN(cmd);
3496 printf("ipfw: opcode %d size truncated\n",
3500 DEB(printf("ipfw: opcode %d\n", cmd->opcode);)
3501 switch (cmd->opcode) {
3513 case O_IPPRECEDENCE:
3530 if (cmdlen != F_INSN_SIZE(ipfw_insn))
3543 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32))
3548 if (cmdlen != F_INSN_SIZE(ipfw_insn_limit))
3553 if (cmdlen != F_INSN_SIZE(ipfw_insn_log))
3556 ((ipfw_insn_log *)cmd)->log_left =
3557 ((ipfw_insn_log *)cmd)->max_log;
3563 /* only odd command lengths */
3564 if ( !(cmdlen & 1) || cmdlen > 31)
3570 if (cmd->arg1 == 0 || cmd->arg1 > 256) {
3571 printf("ipfw: invalid set size %d\n",
3575 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32) +
3580 case O_IP_SRC_LOOKUP:
3581 case O_IP_DST_LOOKUP:
3582 if (cmd->arg1 >= IPFW_TABLES_MAX) {
3583 printf("ipfw: invalid table number %d\n",
3587 if (cmdlen != F_INSN_SIZE(ipfw_insn) &&
3588 cmdlen != F_INSN_SIZE(ipfw_insn_u32))
3593 if (cmdlen != F_INSN_SIZE(ipfw_insn_mac))
3602 if (cmdlen < 1 || cmdlen > 31)
3608 case O_IP_DSTPORT: /* XXX artificial limit, 30 port pairs */
3609 if (cmdlen < 2 || cmdlen > 31)
3616 if (cmdlen != F_INSN_SIZE(ipfw_insn_if))
3621 if (cmdlen != F_INSN_SIZE(ipfw_insn_altq))
3627 if (cmdlen != F_INSN_SIZE(ipfw_insn))
3632 #ifdef IPFIREWALL_FORWARD
3633 if (cmdlen != F_INSN_SIZE(ipfw_insn_sa))
3642 if (ip_divert_ptr == NULL)
3648 if (!NG_IPFW_LOADED)
3652 case O_FORWARD_MAC: /* XXX not implemented yet */
3663 if (cmdlen != F_INSN_SIZE(ipfw_insn))
3667 printf("ipfw: opcode %d, multiple actions"
3674 printf("ipfw: opcode %d, action must be"
3683 if (cmdlen != F_INSN_SIZE(struct in6_addr) +
3684 F_INSN_SIZE(ipfw_insn))
3689 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32) +
3690 ((ipfw_insn_u32 *)cmd)->o.arg1)
3694 case O_IP6_SRC_MASK:
3695 case O_IP6_DST_MASK:
3696 if ( !(cmdlen & 1) || cmdlen > 127)
3700 if( cmdlen != F_INSN_SIZE( ipfw_insn_icmp6 ) )
3706 switch (cmd->opcode) {
3716 case O_IP6_SRC_MASK:
3717 case O_IP6_DST_MASK:
3719 printf("ipfw: no IPv6 support in kernel\n");
3720 return EPROTONOSUPPORT;
3723 printf("ipfw: opcode %d, unknown opcode\n",
3729 if (have_action == 0) {
3730 printf("ipfw: missing action\n");
3736 printf("ipfw: opcode %d size %d wrong\n",
3737 cmd->opcode, cmdlen);
3742 * Copy the static and dynamic rules to the supplied buffer
3743 * and return the amount of space actually used.
3746 ipfw_getrules(struct ip_fw_chain *chain, void *buf, size_t space)
3749 char *ep = bp + space;
3753 /* XXX this can take a long time and locking will block packet flow */
3755 for (rule = chain->rules; rule ; rule = rule->next) {
3757 * Verify the entry fits in the buffer in case the
3758 * rules changed between calculating buffer space and
3759 * now. This would be better done using a generation
3760 * number but should suffice for now.
3765 bcopy(&set_disable, &(((struct ip_fw *)bp)->next_rule),
3766 sizeof(set_disable));
3770 IPFW_RUNLOCK(chain);
3772 ipfw_dyn_rule *p, *last = NULL;
3775 for (i = 0 ; i < curr_dyn_buckets; i++)
3776 for (p = ipfw_dyn_v[i] ; p != NULL; p = p->next) {
3777 if (bp + sizeof *p <= ep) {
3778 ipfw_dyn_rule *dst =
3779 (ipfw_dyn_rule *)bp;
3780 bcopy(p, dst, sizeof *p);
3781 bcopy(&(p->rule->rulenum), &(dst->rule),
3782 sizeof(p->rule->rulenum));
3784 * store a non-null value in "next".
3785 * The userland code will interpret a
3786 * NULL here as a marker
3787 * for the last dynamic rule.
3789 bcopy(&dst, &dst->next, sizeof(dst));
3792 TIME_LEQ(dst->expire, time_uptime) ?
3793 0 : dst->expire - time_uptime ;
3794 bp += sizeof(ipfw_dyn_rule);
3798 if (last != NULL) /* mark last dynamic rule */
3799 bzero(&last->next, sizeof(last));
3801 return (bp - (char *)buf);
3806 * {set|get}sockopt parser.
3809 ipfw_ctl(struct sockopt *sopt)
3811 #define RULE_MAXSIZE (256*sizeof(u_int32_t))
3812 int error, rule_num;
3814 struct ip_fw *buf, *rule;
3815 u_int32_t rulenum[2];
3817 error = suser(sopt->sopt_td);
3822 * Disallow modifications in really-really secure mode, but still allow
3823 * the logging counters to be reset.
3825 if (sopt->sopt_name == IP_FW_ADD ||
3826 (sopt->sopt_dir == SOPT_SET && sopt->sopt_name != IP_FW_RESETLOG)) {
3827 error = securelevel_ge(sopt->sopt_td->td_ucred, 3);
3834 switch (sopt->sopt_name) {
3837 * pass up a copy of the current rules. Static rules
3838 * come first (the last of which has number IPFW_DEFAULT_RULE),
3839 * followed by a possibly empty list of dynamic rule.
3840 * The last dynamic rule has NULL in the "next" field.
3842 * Note that the calculated size is used to bound the
3843 * amount of data returned to the user. The rule set may
3844 * change between calculating the size and returning the
3845 * data in which case we'll just return what fits.
3847 size = static_len; /* size of static rules */
3848 if (ipfw_dyn_v) /* add size of dyn.rules */
3849 size += (dyn_count * sizeof(ipfw_dyn_rule));
3852 * XXX todo: if the user passes a short length just to know
3853 * how much room is needed, do not bother filling up the
3854 * buffer, just jump to the sooptcopyout.
3856 buf = malloc(size, M_TEMP, M_WAITOK);
3857 error = sooptcopyout(sopt, buf,
3858 ipfw_getrules(&layer3_chain, buf, size));
3864 * Normally we cannot release the lock on each iteration.
3865 * We could do it here only because we start from the head all
3866 * the times so there is no risk of missing some entries.
3867 * On the other hand, the risk is that we end up with
3868 * a very inconsistent ruleset, so better keep the lock
3869 * around the whole cycle.
3871 * XXX this code can be improved by resetting the head of
3872 * the list to point to the default rule, and then freeing
3873 * the old list without the need for a lock.
3876 IPFW_WLOCK(&layer3_chain);
3877 layer3_chain.reap = NULL;
3878 free_chain(&layer3_chain, 0 /* keep default rule */);
3879 rule = layer3_chain.reap, layer3_chain.reap = NULL;
3880 IPFW_WUNLOCK(&layer3_chain);
3881 if (layer3_chain.reap != NULL)
3886 rule = malloc(RULE_MAXSIZE, M_TEMP, M_WAITOK);
3887 error = sooptcopyin(sopt, rule, RULE_MAXSIZE,
3888 sizeof(struct ip_fw) );
3890 error = check_ipfw_struct(rule, sopt->sopt_valsize);
3892 error = add_rule(&layer3_chain, rule);
3893 size = RULESIZE(rule);
3894 if (!error && sopt->sopt_dir == SOPT_GET)
3895 error = sooptcopyout(sopt, rule, size);
3902 * IP_FW_DEL is used for deleting single rules or sets,
3903 * and (ab)used to atomically manipulate sets. Argument size
3904 * is used to distinguish between the two:
3906 * delete single rule or set of rules,
3907 * or reassign rules (or sets) to a different set.
3908 * 2*sizeof(u_int32_t)
3909 * atomic disable/enable sets.
3910 * first u_int32_t contains sets to be disabled,
3911 * second u_int32_t contains sets to be enabled.
3913 error = sooptcopyin(sopt, rulenum,
3914 2*sizeof(u_int32_t), sizeof(u_int32_t));
3917 size = sopt->sopt_valsize;
3918 if (size == sizeof(u_int32_t)) /* delete or reassign */
3919 error = del_entry(&layer3_chain, rulenum[0]);
3920 else if (size == 2*sizeof(u_int32_t)) /* set enable/disable */
3922 (set_disable | rulenum[0]) & ~rulenum[1] &
3923 ~(1<<RESVD_SET); /* set RESVD_SET always enabled */
3929 case IP_FW_RESETLOG: /* argument is an int, the rule number */
3931 if (sopt->sopt_val != 0) {
3932 error = sooptcopyin(sopt, &rule_num,
3933 sizeof(int), sizeof(int));
3937 error = zero_entry(&layer3_chain, rule_num,
3938 sopt->sopt_name == IP_FW_RESETLOG);
3941 case IP_FW_TABLE_ADD:
3943 ipfw_table_entry ent;
3945 error = sooptcopyin(sopt, &ent,
3946 sizeof(ent), sizeof(ent));
3949 error = add_table_entry(&layer3_chain, ent.tbl,
3950 ent.addr, ent.masklen, ent.value);
3954 case IP_FW_TABLE_DEL:
3956 ipfw_table_entry ent;
3958 error = sooptcopyin(sopt, &ent,
3959 sizeof(ent), sizeof(ent));
3962 error = del_table_entry(&layer3_chain, ent.tbl,
3963 ent.addr, ent.masklen);
3967 case IP_FW_TABLE_FLUSH:
3971 error = sooptcopyin(sopt, &tbl,
3972 sizeof(tbl), sizeof(tbl));
3975 IPFW_WLOCK(&layer3_chain);
3976 error = flush_table(&layer3_chain, tbl);
3977 IPFW_WUNLOCK(&layer3_chain);
3981 case IP_FW_TABLE_GETSIZE:
3985 if ((error = sooptcopyin(sopt, &tbl, sizeof(tbl),
3988 IPFW_RLOCK(&layer3_chain);
3989 error = count_table(&layer3_chain, tbl, &cnt);
3990 IPFW_RUNLOCK(&layer3_chain);
3993 error = sooptcopyout(sopt, &cnt, sizeof(cnt));
3997 case IP_FW_TABLE_LIST:
4001 if (sopt->sopt_valsize < sizeof(*tbl)) {
4005 size = sopt->sopt_valsize;
4006 tbl = malloc(size, M_TEMP, M_WAITOK);
4011 error = sooptcopyin(sopt, tbl, size, sizeof(*tbl));
4016 tbl->size = (size - sizeof(*tbl)) /
4017 sizeof(ipfw_table_entry);
4018 IPFW_RLOCK(&layer3_chain);
4019 error = dump_table(&layer3_chain, tbl);
4020 IPFW_RUNLOCK(&layer3_chain);
4025 error = sooptcopyout(sopt, tbl, size);
4031 printf("ipfw: ipfw_ctl invalid option %d\n", sopt->sopt_name);
4040 * dummynet needs a reference to the default rule, because rules can be
4041 * deleted while packets hold a reference to them. When this happens,
4042 * dummynet changes the reference to the default rule (it could well be a
4043 * NULL pointer, but this way we do not need to check for the special
4044 * case, plus here he have info on the default behaviour).
4046 struct ip_fw *ip_fw_default_rule;
4049 * This procedure is only used to handle keepalives. It is invoked
4050 * every dyn_keepalive_period
4053 ipfw_tick(void * __unused unused)
4055 struct mbuf *m0, *m, *mnext, **mtailp;
4059 if (dyn_keepalive == 0 || ipfw_dyn_v == NULL || dyn_count == 0)
4063 * We make a chain of packets to go out here -- not deferring
4064 * until after we drop the IPFW dynamic rule lock would result
4065 * in a lock order reversal with the normal packet input -> ipfw
4071 for (i = 0 ; i < curr_dyn_buckets ; i++) {
4072 for (q = ipfw_dyn_v[i] ; q ; q = q->next ) {
4073 if (q->dyn_type == O_LIMIT_PARENT)
4075 if (q->id.proto != IPPROTO_TCP)
4077 if ( (q->state & BOTH_SYN) != BOTH_SYN)
4079 if (TIME_LEQ( time_uptime+dyn_keepalive_interval,
4081 continue; /* too early */
4082 if (TIME_LEQ(q->expire, time_uptime))
4083 continue; /* too late, rule expired */
4085 *mtailp = send_pkt(&(q->id), q->ack_rev - 1,
4086 q->ack_fwd, TH_SYN);
4087 if (*mtailp != NULL)
4088 mtailp = &(*mtailp)->m_nextpkt;
4089 *mtailp = send_pkt(&(q->id), q->ack_fwd - 1,
4091 if (*mtailp != NULL)
4092 mtailp = &(*mtailp)->m_nextpkt;
4096 for (m = mnext = m0; m != NULL; m = mnext) {
4097 mnext = m->m_nextpkt;
4098 m->m_nextpkt = NULL;
4099 ip_output(m, NULL, NULL, 0, NULL, NULL);
4102 callout_reset(&ipfw_timeout, dyn_keepalive_period*hz, ipfw_tick, NULL);
4108 struct ip_fw default_rule;
4112 /* Setup IPv6 fw sysctl tree. */
4113 sysctl_ctx_init(&ip6_fw_sysctl_ctx);
4114 ip6_fw_sysctl_tree = SYSCTL_ADD_NODE(&ip6_fw_sysctl_ctx,
4115 SYSCTL_STATIC_CHILDREN(_net_inet6_ip6), OID_AUTO, "fw",
4116 CTLFLAG_RW | CTLFLAG_SECURE, 0, "Firewall");
4117 SYSCTL_ADD_INT(&ip6_fw_sysctl_ctx, SYSCTL_CHILDREN(ip6_fw_sysctl_tree),
4118 OID_AUTO, "deny_unknown_exthdrs", CTLFLAG_RW | CTLFLAG_SECURE,
4119 &fw_deny_unknown_exthdrs, 0,
4120 "Deny packets with unknown IPv6 Extension Headers");
4123 layer3_chain.rules = NULL;
4124 IPFW_LOCK_INIT(&layer3_chain);
4125 ipfw_dyn_rule_zone = uma_zcreate("IPFW dynamic rule zone",
4126 sizeof(ipfw_dyn_rule), NULL, NULL, NULL, NULL,
4128 IPFW_DYN_LOCK_INIT();
4129 callout_init(&ipfw_timeout, NET_CALLOUT_MPSAFE);
4131 bzero(&default_rule, sizeof default_rule);
4133 default_rule.act_ofs = 0;
4134 default_rule.rulenum = IPFW_DEFAULT_RULE;
4135 default_rule.cmd_len = 1;
4136 default_rule.set = RESVD_SET;
4138 default_rule.cmd[0].len = 1;
4139 default_rule.cmd[0].opcode =
4140 #ifdef IPFIREWALL_DEFAULT_TO_ACCEPT
4145 error = add_rule(&layer3_chain, &default_rule);
4147 printf("ipfw2: error %u initializing default rule "
4148 "(support disabled)\n", error);
4149 IPFW_DYN_LOCK_DESTROY();
4150 IPFW_LOCK_DESTROY(&layer3_chain);
4151 uma_zdestroy(ipfw_dyn_rule_zone);
4155 ip_fw_default_rule = layer3_chain.rules;
4156 printf("ipfw2 (+ipv6) initialized, divert %s, "
4157 "rule-based forwarding "
4158 #ifdef IPFIREWALL_FORWARD
4163 "default to %s, logging ",
4169 default_rule.cmd[0].opcode == O_ACCEPT ? "accept" : "deny");
4171 #ifdef IPFIREWALL_VERBOSE
4174 #ifdef IPFIREWALL_VERBOSE_LIMIT
4175 verbose_limit = IPFIREWALL_VERBOSE_LIMIT;
4177 if (fw_verbose == 0)
4178 printf("disabled\n");
4179 else if (verbose_limit == 0)
4180 printf("unlimited\n");
4182 printf("limited to %d packets/entry by default\n",
4185 error = init_tables(&layer3_chain);
4187 IPFW_DYN_LOCK_DESTROY();
4188 IPFW_LOCK_DESTROY(&layer3_chain);
4189 uma_zdestroy(ipfw_dyn_rule_zone);
4192 ip_fw_ctl_ptr = ipfw_ctl;
4193 ip_fw_chk_ptr = ipfw_chk;
4194 callout_reset(&ipfw_timeout, hz, ipfw_tick, NULL);
4204 ip_fw_chk_ptr = NULL;
4205 ip_fw_ctl_ptr = NULL;
4206 callout_drain(&ipfw_timeout);
4207 IPFW_WLOCK(&layer3_chain);
4208 flush_tables(&layer3_chain);
4209 layer3_chain.reap = NULL;
4210 free_chain(&layer3_chain, 1 /* kill default rule */);
4211 reap = layer3_chain.reap, layer3_chain.reap = NULL;
4212 IPFW_WUNLOCK(&layer3_chain);
4215 IPFW_DYN_LOCK_DESTROY();
4216 uma_zdestroy(ipfw_dyn_rule_zone);
4217 IPFW_LOCK_DESTROY(&layer3_chain);
4220 /* Free IPv6 fw sysctl tree. */
4221 sysctl_ctx_free(&ip6_fw_sysctl_ctx);
4224 printf("IP firewall unloaded\n");